Clim. Past, 8, 1717–1736, 2012 www.clim-past.net/8/1717/2012/ Climate doi:10.5194/cp-8-1717-2012 of the Past © Author(s) 2012. CC Attribution 3.0 License.

A model–data comparison for a multi-model ensemble of early Eocene atmosphere–ocean simulations: EoMIP

D. J. Lunt1, T. Dunkley Jones2,*, M. Heinemann3, M. Huber4, A. LeGrande5, A. Winguth6, C. Loptson1, J. Marotzke7, C. D. Roberts8, J. Tindall9, P. Valdes1, and C. Winguth6 1School of Geographical Sciences, University of Bristol, UK 2Department of Earth Science and Engineering, Imperial College London, UK 3International Pacific Research Center, University of Hawaii, USA 4Department of Earth and Atmospheric Sciences, Purdue University, USA 5NASA/Goddard Institute for Space Studies, USA 6Department of Earth and Environmental Science, University of Texas at Arlington, USA 7Max Planck Institute for Meteorology, Hamburg, Germany 8The Met Office, Exeter, UK 9School of Earth and Environment, University of Leeds, UK *now at: School of Geography, Earth and Environmental Sciences, University of Birmingham, Birmingham, UK

Correspondence to: D. J. Lunt ([email protected])

Received: 29 March 2012 – Published in Clim. Past Discuss.: 16 April 2012 Revised: 19 September 2012 – Accepted: 20 September 2012 – Published: 29 October 2012

Abstract. The early Eocene (∼ 55 to 50 Ma) is a time pe- of atmospheric CO2 and temperature during this time pe- riod which has been explored in a large number of mod- riod will allow a quantitative assessment of the skill of the elling and data studies. Here, using an ensemble of previ- models at simulating warm climates. If it is the case that a ously published model results, making up “EoMIP” – the model which gives a good simulation of the Eocene will also Eocene Modelling Intercomparison Project – and synthe- give a good simulation of the future, then such an assessment ses of early Eocene terrestrial and sea surface temperature could be used to produce metrics for weighting future cli- data, we present a self-consistent inter-model and model– mate predictions. data comparison. This shows that the previous modelling studies exhibit a very wide inter-model variability, but that at high CO2, there is good agreement between models and data for this period, particularly if possible seasonal biases 1 Introduction in some of the proxies are considered. An energy balance analysis explores the reasons for the differences between the Making robust predictions of future climate change is a ma- model results, and suggests that differences in surface albedo jor challenge, which has environmental, societal, and eco- feedbacks, water vapour and lapse rate feedbacks, and pre- nomic relevance. The numerical models which are used to scribed aerosol loading are the dominant cause for the dif- make these predictions are normally tested over time peri- ferent results seen in the models, rather than inconsisten- ods for which there are extensive instrumental records of cies in other prescribed boundary conditions or differences climate available, typically over the last ∼ 100 yr (Hegerl et al., 2007). However, the variations in climate over these in cloud feedbacks. The CO2 level which would give optimal early Eocene model–data agreement, based on those models timescales are small relative to the variations predicted for the next 100 yr or more (Meehl et al., 2007), and so likely which have carried out simulations with more than one CO2 level, is in the range of 2500 ppmv to 6500 ppmv. Given the provide only a weak constraint on future predictions. As spread of model results, tighter bounds on proxy estimates such, proxy indicators of climate from older time periods are increasingly being used to test models. On the timescale of

Published by Copernicus Publications on behalf of the European Geosciences Union. 1718 D. J. Lunt et al.: EoMIP

∼ 100 000 yr, the Paleoclimate Modelling Intercomparison temperature indicators, taking full account of uncertain- Project (PMIP, Braconnot et al., 2007), now in its third phase, ties in the reconstructions. is focusing on three main time periods: the mid-Holocene (6000 yr ago, 6 k), the Last Glacial Maximum (LGM, 21 k), – By analysing the energy balance and fluxes in the mod- and the Last Interglacial (LIG, 125 k). However, these time els, to gain an understanding of the reasons behind the periods are either colder than modern (LGM), or their differences in the model results. warmth is primarily caused not by enhanced greenhouse Section 2 describes the model simulations, Sect. 3 presents gases, but by orbital forcing (mid-Holocene, LIG). As such, the datasets used to evaluate the models, and Sect. 4 presents their use for testing models used for future climate prediction the model results and model–data comparison. Section 5 is also limited. On the timescale of millions of years, sev- quantifies the reasons for the differences between the model eral time periods show potential for model evaluation, being results, and Sect. 6 discusses, concludes, and proposes direc- characterised by substantial warmth which is thought to be tions for future research. driven primarily by enhanced atmospheric CO2 concentra- tions. An example is the mid-Pliocene (3 million years ago, 3 Ma), when global annual temperature was ∼ 3 ◦C greater 2 Model simulation descriptions than pre-industrial (Dowsett et al., 2009). However, the lat- est estimates of mid-Pliocene CO2 (Pagani et al., 2010; Seki Many model simulations have been carried out over the last et al., 2010) range from ∼ 360 to ∼ 420 ppmv, which is sim- two decades with the aim of representing the early Eocene. ilar to that of modern (∼ 390 ppmv in 2010 according to the Here, we present and discuss results from a selection of Scripps CO2 program, http://scrippsco2.ucsd.edu/), and sub- these. We present all simulations of which we are aware that stantially less than typical IPCC scenarios for CO2 concen- (a) are published in the peer-reviewed literature, and (b) are tration at the end of this century (∼ 1000 ppmv in the A1F1 carried out with fully dynamic atmosphere–ocean general scenario, > 1370 ppmv CO2-equivalent in the RCP8.5 sce- circulation models (GCMs), with primitive equation atmo- nario, Meehl et al., 2007; Moss et al., 2010). The time period spheres. This makes a total of 4 models – HadCM3L (Lunt which shows possibly the most similarity to projections of et al., 2010), ECHAM5/MPI-OM (Heinemann et al., 2009), the end of the 21st century and beyond is the early Eocene, CCSM3 (Winguth et al., 2010, 2012; Liu et al., 2009; Huber ∼ 55 to ∼ 50 Ma. A recent compilation of Cenozoic atmo- and Caballero, 2011), and GISS ModelE-R (Roberts et al., spheric CO2 is relatively data-sparse during the early Eocene, 2009). Criterion (b) is chosen to select the models which are with large uncertainty range, meaning that values more than most similar to those used in future climate change projection 2000 ppmv cannot be ruled out (Beerling and Royer, 2011). (i.e. we exclude models with energy balance atmospheres Relatively high values for the early Eocene are consistent such as GENIE; Panchuk et al., 2008). There are two sets with recent CO2 reconstructions for the latest Eocene of the of CCSM3 simulations, which we name CCSM W(Winguth order 1000 ppmv (Pearson et al., 2009; Pagani et al., 2011). et al., 2010, 2012) and CCSM H(Liu et al., 2009; Huber Proxy indicators have been interpreted as showing tropical and Caballero, 2011). All the models and simulations are temperatures at this time ∼5 ◦C warmer than modern (e.g. summarised in Table 1. Together they make up the “Eocene Pearson et al., 2001), and high latitude terrestrial tempera- Modelling Intercomparison Project”, EoMIP. EoMIP differs tures more than 20 ◦C warmer (e.g. Huber and Caballero, from more formal model intercomparisons, such as those 2011). Recently, due at least in part to interest associated carried out under the auspices of PMIP, in that the groups with this time period as a possible future analogue, there have carried out their own experimental design and simula- have been a number of new sea surface temperature (SST) tions in isolation, and the comparison is being carried out and terrestrial temperature data published, using a range of post hoc, rather than being planned from the outset. As such, proxy reconstruction methods. There have also been several the groups have used different paleogeographical boundary models recently configured for the early Eocene and attempts conditions and CO2 levels to simulate their Eocene climates. made to understand the mechanisms of Eocene warmth. Most This has advantages and disadvantages compared to the more of these studies have carried out some form of model–data formal approach with a single experimental design: The main comparison; however, the models have not been formally in- disadvantage is that a direct comparison between models is tercompared in a consistent framework, and new data now impossible due to even subtle differences in imposed bound- allows a more robust and extensive evaluation of the models. ary conditions; the main advantage is that in addition to un- The aims of this paper are: certainties in the models themselves, the model ensemble also represents the uncertainties in the paleoenvironmental – To present an intercomparison of five models, all re- conditions, and therefore more fully represents the uncer- cently used to simulate the early Eocene climate. tainty in our climatic predictions for that time period.

– To carry out a consistent and comprehensive com- parison of the model results with the latest proxy

Clim. Past, 8, 1717–1736, 2012 www.clim-past.net/8/1717/2012/ D. J. Lunt et al.: EoMIP 1719

Table 1. Summary of model simulations in EoMIP. Some models have irregular grids in the atmosphere and/or ocean, or have spectral atmospheres. The atmospheric and ocean resolutions are given in number of grid boxes, X × Y × Z where X is the effective number of grid boxes in the zonal, Y in the meridional, and Z in the vertical. See the original references for more details.

Name Eocene simulation reference Model name and reference Atmosphere resolution Ocean resolution HadCM Lunt et al. (2010) HadCM3L, Cox et al. (2001) 96 × 73 × 19 96 × 73 × 20 ECHAM Heinemann et al. (2009) ECHAM5/MPI-OM, Roeckner 96 × 48 × 19 142 × 82 × 40 et al. (2003) CCSM W Winguth et al. (2010, 2012) CCSM3, Collins et al. (2006); 96 × 48 × 26 100 × 116 × 25 Yeager et al. (2006) CCSM H Liu et al. (2009) CCSM3, Collins et al. (2006); 96 × 48 × 26 100 × 122 × 25 Huber and Caballero (2011) Yeager et al. (2006) GISS Roberts et al. (2009) GISS ModelE-R, Schmidt et al. 72 × 45 × 20 72 × 45 × 13 (2006)

Name Paleogeography Sim. length CO2 levels Vegetation Aerosols (years) HadCM proprietary > 3400 ×2,4,6 homogenous shrubland as control ECHAM Bice and Marotzke (2001) 2500 ×2 homogenous woody savanna as control CCSM W Sewall et al. (2000) 1500 ×4,8,16 Shellito and Sloan (2006) as control with marginal sea parameterisation CCSM H Sewall et al. (2000) > 3500 ×2,4,8,16 Sewall et al. (2000) reduced aerosol GISS Bice and Marotzke (2001) 2000 ×2 Sewall et al. (2000) as control

2.1 HadCM we use in this paper in Sect. 5). They reported a larger polar warming than many previous studies, which they attributed to Lunt et al. (2010) investigated the potential role of hydrate local radiative forcing changes rather than modified poleward destabilisation as a mechanism for the Paleocene–Eocene heat transport. The Eocene simulation was carried out under ∼ Thermal Maximum (PETM, 55 Ma), using the HadCM3L ×2 CO2 levels, and a globally homogeneous vegetation was model. They found a switch in modelled ocean circulation prescribed, with characteristics similar to present-day woody which occurred between ×2 and ×4 pre-industrial concen- savanna. trations of atmospheric CO2, which resulted in a non-linear warming of intermediate ocean depths. They hypothesised 2.3 CCSM W and CCSM H that this could be a triggering mechanism for hydrate release. For the three Eocene simulations carried out (×2, ×4, and Huber and Caballero (2011) presented a set of Eocene ×6), vegetation was set globally to a “shrub” plant functional CCSM3 simulations, originally published by Liu et al. type. The paleogeography is proprietary but is illustrated in (2009), with the main aim of comparing these with a new Supplementary Information of Lunt et al. (2010). An addi- compilation of proxy terrestrial temperature data. They found × tional simulation at ×6 CO2 was carried out with the same that at high CO2 ( 16) they obtained good agreement with model by Tindall et al. (2010), which incorporated oxygen data from mid and high latitudes. We use this same proxy isotopes into the hydrological cycle. The δ18O of seawater dataset in this paper, including estimates of uncertainty, for from the Tindall et al. (2010) simulation is used in our SST evaluating all the EoMIP simulations. compilation to inform the uncertainty range of the proxies Winguth et al. (2010) and Winguth et al. (2012) carried based on δ18O measurements (see Sect. 3). out an independent set of CCSM3 simulations motivated by investigating the role of hydrates as a possible cause of the 2.2 ECHAM PETM. They found evidence of non-linear ocean warming and enhanced stratification in response to increasing atmo- Heinemann et al. (2009) presented an ECHAM5/MPI-OM spheric CO2 concentrations, and a shift of deep water forma- Eocene simulation and compared it to a pre-industrial sim- tion from northern and southern sources to a predominately ulation, diagnosing the reasons for the Eocene warmth by southern source. making use of a simple 1-D energy balance model (which www.clim-past.net/8/1717/2012/ Clim. Past, 8, 1717–1736, 2012 1720 D. J. Lunt et al.: EoMIP

The CCSM W and CCSM H simulations differ mainly in estimates for sea surface (archaeal-derived isoprenoid glyc- the treatment of aerosols. In the CCSM W simulation, a high erol dibiphytanyl glycerol tetraether (GDGT) paleothermom- aerosol load is applied, whereas the CCSM H simulation etry), near-sea surface (mixed layer dwelling planktonic considers a lower-than-present-day aerosol distribution fol- foraminifera) and shallow inner-shelf bottom waters (bivalve lowing the approach by Kump and Pollard (2008), possibly oxygen isotopes) from across the early Eocene (Ypresian justified by a reduced ocean productivity and thus reduced stage; ∼ 55.9 to 49 Ma). Also included in the compilation are DMS (Dimethyl sulfide) emissions. A globally reduced pro- some data from the very latest Paleocene, within the interval ductivity is supported by the recent study of Winguth et al. immediately before but not including the Paleocene–Eocene (2012). However, it remains uncertain to what extent inten- Thermal Maximum (PETM). These data are included to in- sified volcanism near the PETM might have increased the crease the geographical coverage of data, especially in the aerosol load (Storey et al., 2007). mid to low latitudes. Long-term paleotemperature records through the early 2.4 GISS Eocene indicate the presence of a significant warming trend in both oceanic intermediate waters of ∼4 ◦C(Zachos et al., Roberts et al. (2009) carried out an investigation into the role 2008) and high-latitude sea surface temperatures of up to of the geometry of Arctic gateways in determining Eocene ∼ 10 ◦C(Bijl et al., 2009; Hollis et al., 2012) in the lead-up to climate with the GISS ModelE-R, configured with ×4 CO2 the early Eocene Climatic Optimum (EECO) (Zachos et al., and ×7 CH4 compared with pre-industrial. They estimated 2008). Although tropical sea surface temperatures may have the change to the total forcing to be about ×4.3 of CO2- been more stable (Pearson et al., 2007) through this interval, equivalent, but for the purposes of this paper we assume their attention should be paid to the age of the compiled proxy simulations were at ×4 CO2. They found that restricting Arc- records relative to this warming trend. Notably, comparison tic gateways led to warming of the North Atlantic and fresh- of EECO records from the southern high latitudes with basal ening of the Arctic Ocean, similar to data associated with early Eocene records from the mid and low latitudes, or non- the “Azolla” event (Brinkhuis et al., 2006). They incorpo- EECO proxy records from the high latitudes, could introduce rated oxygen isotopes into the hydrological cycle in their a spurious reduction in the latitudinal temperature gradient or model (Roberts et al., 2011), and used the predicted isotopic false high latitude proxy–proxy disagreements, respectively. concentrations of seawater to more directly compare with The relative paucity of the available data however, which is 18 proxy temperature estimates. The δ O of seawater from the taken from a small number of locations, many of which have Roberts et al. (2011) simulation is used in our SST compila- limited time series and/or poor age control, prohibits a nar- tion to inform the uncertainty range of the proxies based on rowly focused time-slice reconstruction of SSTs within the 18 δ O measurements (see Sect. 3). early Eocene. Instead, we have chosen to divide the data into two broad categories, those from the period of peak Cenozoic warmth during the EECO, and the remainder, assigned to a 3 Early Eocene SST and land temperature datasets generally cooler “background” early Eocene climate state. To evaluate the various climate model simulations, we make Pre-PETM records are included in this “background” early use of both terrestrial and marine temperature datasets. The Eocene compilation. Given there is some evidence for warm- marine dataset has been compiled for this paper, and the ter- ing between pre- and post-PETM conditions in the high lati- restrial data is identical to that presented in Huber and Ca- tudes (Bijl et al., 2009; Hollis et al., 2012), as well as the gen- ballero (2011). In both cases we take as full account as pos- eral early Eocene warming trend, these pre-PETM records sible of the various uncertainties associated with each proxy. are likely to have a slight cool bias relative to other estimates The purpose of these compilations is not to provide a of early Eocene temperatures. tightly constrained “time-slice” reconstruction of any point The identification of EECO records is only possible where in the early Eocene against which the ensemble or individ- there is either good age control and/or a clear temperature- ual model runs can be compared; instead, we include data trend across a long-term early Eocene record. Only three ma- spanning the entire early Eocene. This approach is consis- rine SST proxy datasets are identified as representing EECO tent with the EoMIP simulations themselves, in which mod- conditions for all or part of the associated SST time series – els have not been run with the same specific set of simulation ODP Site 1172D, Waipara River and the Arctic IODP Site boundary conditions, such as paleogeography or atmospheric M0004. The EECO in the ODP Site 1172D record is iden- greenhouse gas forcings, but can be considered to reflect a tified following Hollis et al. (2012) as spanning ∼ 53.1 to possible range of time periods within the early Eocene. ∼ 49 Ma (588.85 to 562.70 m b.s.f.), with the pre-EECO in- terval from 54.9 to 53.3 Ma (611.0 to 591.15 m b.s.f.). All of 3.1 Marine dataset the Waipara River data used here are identified as represent- ing EECO conditions (Hollis et al., 2009, 2012). Although For the purposes of model–data comparison, we have the early Eocene age model for the Arctic IODP M0004 compiled (see Supplement) available paleotemperature is poorly constrained between early Eocene hyperthermal

Clim. Past, 8, 1717–1736, 2012 www.clim-past.net/8/1717/2012/ Lunt et al: EoMIP 15 D. J. Lunt et al.: EoMIP 1721 events and the termination of the “Azolla” phase, there is Proxy temperatures [degrees C] 50 a distinct cooling in GDGT-derived proxy temperature es- mean SST Mg/Ca sw=3 d18O Zachos TEX86_H Mg/Ca sw=4 d18O Tindall 1/TEX86 timates between the stratigraphically lower (core 27X) and Mg/Ca sw=5 d18O Roberts TEX86_L upper (cores 23X to 19X) parts of this section (Sluijs et al., 40 2008). For the purposes of a more refined proxy–proxy and proxy–model comparison, we have labelled the (warmer) 30 SST data from the lower part of the M0004 as pre-EECO and the (cooler) SST data from the upper part as EECO. This is in line with speculation in Sluijs et al. (2008) that 20 temperature while there is a global trend of warming through the early Eocene, Arctic SSTs reduced during this interval. Data from 10 M0004 Core 27X between 369 to 367.9 rmcd has also been excluded, which, based on carbon isotope stratigraphy, likely 0 represents the early Eocene hypethermal event ETM2 (Sluijs et al., 2008). -50 0 50 latitude The remaining sites have either relatively poorly con- strained age models and have no discernable SSTFig. trend1. SST proxy dataset,Fig. compiled1. SST for proxy this study. dataset, Coloured compiled symbols show for the this various study. median Coloured estimates from sym- the literature, with various 18 through the dataset used (ODP Sites 690 and 738),asumptions are spot about Mg/Cabols of show sewater, theδ Osw various, and TEX median86 calibration. estimates Error bars from indicate the the maximum literature, and minimum with range at each site including temporal variability and calibration error. The filled black circles represent the mean18 SST at each site, averaged over the various samples within the early Eocene (Tanzania, Hatchetigbeeassumptions. Largervarious symbols represent assumptions ‘background’ about early Mg/Ca Eocene state, ofseawater, smaller symbolsδ representOsw, and the EECO. TEX See86 Section 3.1 for more Bluff) or are well-constrained pre-PETM records. Seymourdetails. calibration. Error bars indicate the maximum and minimum range Island is the exception to this, where there remains uncer- at each site including temporal variability and calibration error. tainty about even the gross age of this succession. The data The filled black circles represent the mean SST at each site, aver- used in this compilation is sourced from Telms 3 to 5, which, aged over the various assumptions. Larger symbols represent “back- ground” early Eocene state, smaller symbols represent the EECO. based on strontium isotope stratigraphy and sparse biostrati- See Sect. 3.1Globalfor mean SST more [degrees C] details. Global mean surface temperature [degrees C] graphic data, were thought to span the early Eocene, ex- 30 CCSM_H 30 CCSM_H 25 tending just across the early/middle Eocene boundary (Ivany CCSM3_W GISS CCSM3_W ECHAM5 20 HadCM3L et al., 2008). A revised age assessment, based on dinoflag- non-analogue behaviourGISS HadCM3L is very difficult to assess and we 25 ECHAM5 ellate biostratigraphy, suggests that the lower part of this 15 dotemperature not try to quantify this directlytemperature in our uncertainty anal-

10 sequence is middle and not early Eocene in age (Dou- GISS 20 ysis. We do, however, attempt to quantifyECHAM5NCEP uncertainty asso- HadCM3LCCSM3_WCCSM_H HADISSTGISS 5 glas et al., 2011). This new biostratigraphic data remains ciatedHadCM3LECHAM5 with both paleotemperature calibrations and the es- CCSM3_WCCSM_H

15 0 somewhat tentative, and while awaiting the publication of x1 x2 x4 x8 x16 x1 x2 x4 x8 x16 timates of ancientCO2 relative to preindustrial seawater chemistry. ThisCO2 relative is to preindustrial achieved by a fully revised age model for these successions, the Sey- (1) applying(a) the standard error determined(b) from the mod- mour Island data is provisionally included in the SST com- ern calibration dataset to paleotemperature estimates; (2) the pilation. It is however, assigned to the backgroundFig. 2.“pre-Global annualuse mean of(a) SST multiple (hSST i) and alternate (b) continental calibrations 2m air temperature where (hLAT i), there as a function is ofongo- CO2 for all simulations, and for observational datasets. The simulations at ×1 CO are pre-industrial reference simulations. EECO” category, as it appears to be more likely to be rep- ing debate about the2 most appropriate calibration or proxy resentative of middle Eocene post-EECO cooling. It is also method (GDGT paleothermometry) or where modern cal- noted that although Tanzanian Drilling Project Site 2 (TDP2) ibrations vary with environmental conditions (oxygen iso- was originally reported as extending down into the early topes); and (3) applying multiple estimates (Mg/Ca) or dif- Eocene (Nicholas et al., 2006; Pearson et al., 2004), with ferent estimation methodologies (oxygen isotopes) for sea- the resolution of planktic foraminifera–nannofossil biostrati- water chemistry. Where distinct proxies (GDGT paleother- graphic mismatches around the early/middle Eocene bound- mometry) or distinct parameters for seawater chemistry are ary (Payros et al., 2007), TDP2 is now considered to be used (Mg/Ca and oxygen isotopes), the derived temperature entirely within the basal middle Eocene (P. Pearson, per- ranges are calculated separately at each site. This leads to sonal communication, July 2012) and is excluded from this H L the following sets of proxy data: TEX86, TEX86, 1/TEX86, compilation. oxygen isotope paleothermometry with modelled and latitu- 18 For each location the primary geochemical proxy data dinal corrected δ Osw estimates, and Mg/Ca paleothermom- −1 were first collated and then used to generate a range of SST etry with assumed Mg/Casw values of 3, 4 and 5 mol mol . estimates based on a set of plausible assumptions about the Details of all these methods and calculations are described underlying paleotemperature methodology. All of the pale- below, and illustrated in Fig. 1. otemperature estimation methods are subject to uncertainty arising from their present-day calibrations, necessary as- 3.1.1 Oxygen isotopes sumptions about ancient seawater chemistry and potential non-analogue behaviour between modern and ancient sys- For planktic foraminifera-derived δ18O temperature esti- tems. Although positive steps are being made with deep-time mates, we applied the temperature–δ18O calibrations of Be- proxy inter-comparison studies (Hollis et al., 2012), potential mis et al. (1998) for the symbiotic planktic foraminifera, www.clim-past.net/8/1717/2012/ Clim. Past, 8, 1717–1736, 2012 1722 D. J. Lunt et al.: EoMIP

Orbulina universa, under both high and low light con- for biogenic aragonite, to be ±1.4 ◦C. A SMOW to VPDB ditions (their Eqs. 1 and 2). Together, these two equa- correction of −0.2 is already implicit within the Kobashi tions bracket most of the natural variability in planktic et al. (2003) form ofh this temperature equation. foraminifera temperature–δ18O space within modern plank- ton tow data (Bemis et al., 1998). Unlike the multiple GDGT 3.1.2 Mg/Ca ratios of planktonic foraminifera paleotemperature equations, which seem to vary in their ac- curacy with geographical location or paleoenvironment, the To estimate calcification temperature, we used the multi- two equations of Bemis et al. (1998) represent the natural species sediment trap calibration of Anand et al. (2003), variability at a single location (high/low light conditions). which has a calibration standard deviation of ±1.13 ◦C. The derived SST estimates from these two equations are thus This paleotemperature estimation relies strongly upon the combined into a single range representing the potential en- assumed value of the Mg/Ca ratio in early Eocene seawa- vironmental variability at any location. The standard errors ter, which is still poorly constrained. Values in the range of on Eq. (1) (low light) and 2 (high light) are ±0.7 and 0.5 ◦C, 3–4 mol mol−1 are typically used within paleoceanographic respectively. studies, and these produce tropical (Sexton et al., 2006) and Three sets of temperature estimate are, however, plotted mid-latitude (Creech et al., 2010) surface ocean tempera- in Fig. 1 for each location with planktic foraminifera δ18O tures that are broadly consistent with independent paleotem- 18 data based on three estimates of δ Osw: the latitudinal cor- perature estimates. Here, we separately calculate and plot rection of Zachos et al. (2008) and the modelled mixed-layer (in Fig. 1) paleotemperature estimates based on three val- 18 δ Osw of Tindall et al. (2010) and of Roberts et al. (2011). ues of seawater Mg/Ca across a wide range, namely 3, 4 and The latitudinal correction is a first-order approximation of the 5 mol mol−1. Distinguishing temperature estimates based on 18 effects of the global hydrological cycle on seawater δ Osw these three values allows for (1) a clear representation of the and is a widely used improvement on an “ice-free” globally sensitivity of temperature to the assumed value of Mg/Casw, 18 uniform estimate of δ Osw. This empirical relationship does and (2) the future use of the most appropriate temperature not, however, include zonal deviations from this general pat- range if/when more robust constraints on early Paleogene tern. In the early Paleogene world, these zonal deviations are Mg/Casw become available. likely to have been significantly different to the modern due For reference, an estimate of ∼ 3.5 mol mol−1 is ob- to the closure of the major Southern Ocean gateways and tained using the Lear et al. (2002) calibration for Oridor- the resulting high-latitude isolation of the Atlantic and Pa- salis umbonatus, the paired foraminifera Mg/Ca value of cific basins. The isotope-enabled versions of HadCM3L and 2.78 mol mol−1 and δ18O-derived bottom water temperature GISS, however, reproduce both the expected latitudinal gra- of 12.4 ◦C they quote for ∼ 49 Ma. A lower, ∼ 3 mol mol−1 18 dients in δ Osw and an estimation of early Paleogene basin- value is obtained by the same method but using the re- to-basin zonal gradients (Tindall et al., 2010; Roberts et al., vised calibrations for O. umbonatus (Rathmann et al., 2004; 18 2011). These modelled δ Osw values, taken from modelled Rathmann and Kuhnert, 2008). Higher values of ∼ 4 to ocean depths of 50 m, are a potential refinement to early Pa- 5 mol mol−1 are indicated by δ18O–Mg/Ca paleotempera- 18 leogene δ Osw estimation and are included here to provide ture inter-comparisons with well-preserved early Eocene a more comprehensive assessment of the range of possible foraminifera (Sexton et al., 2006), whilst recent modelling δ18O SST estimates. of trace metal fluxes and assessments of the long-term ben- For the latitudinal correction, we assume a global average thic foraminifera δ18O–Mg/Ca record suggest values of ∼ δ18O composition of early Eocene seawater of −1 to be 3 mol mol−1 or less (Cramer et al., 2011; Farkasˇ et al., consistent with the value used by Tindall et al. (2010).h This is 2007). These lower values are more consistent with estimates comparable to the −0.96 used in a recent early Eocene pa- based on ridge flank hydrothermal carbonate veins, at around −1 leotemperature inter-comparisonh (Hollis et al., 2012), and the 2 mol mol (Coggon et al., 2010). There remains a pressing value of −0.81 used in Roberts et al. (2011). A SMOW to need to understand the causes of these discrepancies and es- VPDB conversionh of −0.27 was applied to all estimates of tablish robust estimates of the Mg/Ca ratio of ancient seawa- 18 δ Osw. The maximum differenceh in paleotemperature esti- ter (see discussion in Coggon et al., 2011). 18 mates between the three δ Osw assumptions is for Seymour 18 Island, where the median value using the modelled δ Osw 3.1.3 TEX86 of Roberts et al. (2011) is 5.6 ◦C warmer than that using the 18 δ Osw of Tindall et al. (2010). Determining the appropriate proxy index and calibration For the Eurhomalea and Cucullaea bivalve δ18O data, for deep-time paleotemperature estimates based on the we used the biogenic aragonite δ18O–temperature calibration relative abundances of archaeal-derived isoprenoid glyc- of Grossman and Ku (1986) as modified by Kobashi et al. erol dibiphytanyl glycerol tetraethers (GDGTs) is problem- 18 (2003) with both the latitude-corrected and modelled δ Osw atic. Three methods have recently been proposed: separate noted above. We calculated the standard error on the Gross- “low” and “high” temperature proxies based on different L H man and Ku (1986) calibration, based on the original dataset ratios of GDGTs, TEX86 and TEX86 (Kim et al., 2010),

Clim. Past, 8, 1717–1736, 2012 www.clim-past.net/8/1717/2012/ Lunt et al: EoMIP 15

Proxy temperatures [degrees C] 50 mean SST Mg/Ca sw=3 d18O Zachos TEX86_H Mg/Ca sw=4 d18O Tindall 1/TEX86 Mg/Ca sw=5 d18O Roberts TEX86_L 40

30

20 temperature

10

0

-50 0 50 latitude

Fig. 1. SST proxy dataset, compiled for this study. Coloured symbols show the various median estimates from the literature, with various 18 asumptions about Mg/Ca of sewater, δ Osw, and TEX86 calibration. Error bars indicate the maximum and minimum range at each site including temporal variability and calibration error. The filled black circles represent the mean SST at each site, averaged over the various assumptions. Larger symbols represent ‘background’ early Eocene state, smaller symbols represent the EECO. See Section 3.1 for more details.

D. J. Lunt et al.: EoMIP 1723

Global mean SST [degrees C] Global mean surface temperature [degrees C] 30

CCSM_H 30 CCSM_H 25 CCSM3_W GISS CCSM3_W ECHAM5 20 HadCM3L GISS HadCM3L

25 ECHAM5 15 temperature temperature

10 20 GISS ECHAM5NCEP HadCM3LCCSM3_WCCSM_H HADISSTGISS 5 HadCM3LECHAM5 CCSM3_WCCSM_H

15 0 x1 x2 x4 x8 x16 x1 x2 x4 x8 x16 CO2 relative to preindustrial CO2 relative to preindustrial

(a) (b)

Fig. 2. Global annual mean (a) SST (hSSTi) and (b) continental 2 m air temperature (hLATi), as a function of CO2 for all simulations, and for observational datasets. The simulations at ×1 CO2 are pre-industrial reference simulations. Fig. 2. Global annual mean (a) SST (hSST i) and (b) continental 2m air temperature (hLAT i), as a function of CO2 for all simulations, and for observational datasets. The simulations at ×1 CO2 are pre-industrial reference simulations.

and a non-linear calibration of the original TEX86 index, to include this published data but note these higher BIT in- “1/TEX86”(Liu et al., 2009), revised by Kim et al. (2010). dices. In some cases, as for the early Eocene data from Tanza- H L TEX86 and 1/TEX86 are based on the same underlying ratio nia, TEX86 temperature estimates are clearly erroneous and of GDGTs – the original TEX86 proxy – but differ in the form are excluded. Due to the greater availability of data, sam- of their calibration equations (logarithmic versus reciprocal). ples from the Arctic Ocean IODP Site M0004 with BIT in- The fundamentally different ratio of GDGT isomers within dices > 0.3 were excluded. The standard TEX86 proxies dis- L TEX86 results in a proxy that can, in certain instances, pro- cussed above can be applied to this early Eocene Arctic data H duce temperature trends contrary to TEX and 1/TEX86. It rather than the TEX86 proxy used through the PETM by 86 ◦ is, as yet, unclear which of these proxies is the most appro- Sluijs et al. (2006). The errors ( C) on the GDGT-based prox- ± H ± L priate for early Eocene paleotemperature estimation. There ies are 2.5 for TEX86 (GDGT index-2), 4.0 for TEX86 are indications that their suitability may vary with both the (GDGT index-1) and ±5.4 for 1/TEX86 (Kim et al., 2010). temperature range and paleoenvironment of GDGT forma- From the arrays of time-varying temperature estimates at tion (Hollis et al., 2012). each site, and for all assumptions of seawater composition For the purposes of this study, we separately calculate and TEX86 calibration, we calculated the median, maximum and plot (in Fig. 1) paleotemperatures at each site using all and minimum values from the time series as the basis for the H L model–data comparisons. There is an important caveat to this three measures: TEX86, TEX86 and 1/TEX86. This illustrates the full range of temperature estimates produced by these approach that relates to the effect of data quantity and strati- GDGT-based proxies at a given location but also allows for a graphic range on the temperature envelopes plotted. Where more refined use of this data as the behaviour of these prox- there are reasonably extensive time series, natural tempo- ies becomes better understood. The calculation of all three ral variability can result in a larger envelope of temperature proxies at all sites may be considered by some to be an estimates than at sites where data is limited to a few spot erroneous application of, for example, a “low temperature” samples. As a result, these envelopes should not be taken to L solely represent uncertainty in paleotemperature estimation, proxy, TEX86, to the mid and low latitudes of a warm cli- mate state. We must stress that we do not intend to imply but to also include a measure of the temporal variability at that all three are equally applicable at all sites. Rather, by individual sites. showing all three proxies at all sites alongside other proxy Finally, in order to provide a single “zero-order estimate” temperature estimates, we hope to contribute to the ongoing for SST at each location, we averaged the median esti- discussion about the behaviour of GDGT proxies in deep- mates at each site across the various assumptions of seawater time paleoenvironments (Hollis et al., 2012). chemistry and TEX86 calibration. The resulting estimates are Recent development of good practice suggests the exclu- shown as filled black circles in Fig. 1, separately for EECO sion of paleotemperature estimates from samples with a BIT and non-EECO estimates. We are not suggesting that these index in excess of 0.3 (Kim et al., 2010). Although we ac- are “best” estimates – interpretation of proxies is undergoing cept this as a recommendation, in the existing published data rapid development at present – instead their main use is to compiled here it would result in the exclusion of all early provide single-value estimates in the maps in Figs. 3 and 9a, Eocene data from Tanzania and Hatchetigbee Bluff, which and to provide a zero-order target for the model error scores both have BIT indices in the range 0.3 to 0.5. We choose in Table 2. By providing the raw data in the Supplement,

www.clim-past.net/8/1717/2012/ Clim. Past, 8, 1717–1736, 2012 16 Lunt et al: EoMIP

1724 D. J. Lunt et al.: EoMIP

×2 ×4 ×6 ×8 ×16

HadCM3L - 2* CO2 anomaly HadCM3L - 4* CO2 anomaly HadCM3L - 6* CO2 anomaly

90N 90N 90N

60N 60N 60N

30N 30N 30N

0 0 0

30S 30S 30S

60S 60S 60S

90S 90S 90S HadCM 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

ECHAM5 - 2* CO2 anomaly

90N

60N

30N

0

30S

60S

90S ECHAM 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

CCSM3_W - 4* CO2 anomaly CCSM3_W - 8* CO2 anomaly CCSM3_W - 16* CO2 anomaly

90N 90N 90N

60N 60N 60N

30N 30N 30N

0 0 0

30S 30S 30S

60S 60S 60S

90S 90S 90S CCSM W 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

CCSM_H - 2* CO2 anomaly CCSM_H - 4* CO2 anomaly CCSM_H - 8* CO2 anomaly CCSM_H - 16* CO2 anomaly

90N 90N 90N 90N

60N 60N 60N 60N

30N 30N 30N 30N

0 0 0 0

30S 30S 30S 30S

60S 60S 60S 60S

90S 90S 90S 90S CCSM H 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

GISS - 4* CO2 anomaly

90N

60N

30N

0

30S

60S

90S GISS 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

-20.0 -16.0 -12.0 -8.0 -4.0 0.0 4.0 8.0 12.0 16.0 20.0 Temperature (Celsius)

m m Fig. 3. SST anomaly in the model simulations (SSTe −SSTp ), as a function of model and fractional CO2 increase from pre-industrial. Also d d m m Fig. 3.shownSST for anomaly the proxies in the are model SSTe − simulationsSSTp. (SSTe −SSTp ), as a function of model and fractional CO2 increase from pre-industrial. d d Also shown for the proxies are SSTe −SSTp .

others will be able to provide their own interpretation if nec- 4 Results and model–data comparison essary as data is re-interpreted. In this section, we present results from the EoMIP model 3.2 Terrestrial dataset ensemble (early Eocene simulations and pre-industrial con- trols), as described in Sect. 2, and compare them with the ×2 ×4 ×6 ×8 ×16

HadCM3L - 2* CO2 anomaly HadCM3L - 4* CO2 anomaly HadCM3L - 6* CO2 anomaly For the terrestrial, we make use of the data compilation pre- 90N 90N 90N data described in Sect. 3. The reasons for the different model

60N 60N 60N

sented in Huber and Caballero30N (2011). This is30N based largely 30N results are explored in more detail in Sect. 5.

0 0 0 on macrofloral assemblages, with30S mean annual30S temperatures 30S It is useful at this stage to define some nomenclature. To 60S 60S 60S

90S 90S 90S being reconstructedHadCM primarily180W by150W 120W 90W leaf-margin60W 30W 0 30E 60E 90E 120E 150E 180E analysis180W 150W 120W 90W 60W 30W 0 30E and/or60E 90E 120E 150E 180E 180W 150W 120W 90Wrepresent60W 30W 0 30E 60E 90E 120E 150E 180E the distribution of temperature, we write SST for ECHAM5 - 2* CO2 anomaly

90N

CLAMP (physiognomic analysis60N of leaf fossils). Other prox- sea surface temperature (only defined over ocean), or LAT

30N ies are also incorporated, such0 as isotopic estimates, organic for land near-surface (∼ 1.5 m) air temperature (only defined 30S

geochemical indicators, and palynoflora.60S The error bars asso-

90S over continents), or GAT for near-surface air temperature ECHAM 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

CCSM3_W - 4* CO2 anomaly CCSM3_W - 8* CO2 anomaly CCSM3_W - 16* CO2 anomaly

ciated with each data point incorporate uncertainty90N in calibra- (defined globally),90N or GST for90N surface temperature (defined 60N 60N 60N tion, topography, and dating. More information30N on the data globally), or30N just T for a generic30N temperature. Global means 0 0 0 themselves, and the estimates of uncertainty, can30S be found in 30S 30S 60S are denoted60S by angled brackets,60S so that, for example, the

90S 90S 90S Huber and CaballeroCCSM(2011W). 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E CCSM_H - 2* CO2 anomaly CCSM_H - 4* CO2 anomaly CCSM_H - 8* CO2 anomaly CCSM_H - 16* CO2 anomaly h i global mean sea surface temperature is SST . Zonal means 90N 90N 90N 90N

60N 60N 60N 60N

30N 30N are denoted30N by overbars, so30N that the zonal mean sea surface

Both marine and terrestrial0 datasets are provided0 in the 0 0

30S 30S temperature30S is SST. In the30S case of model output, ensem-

Supplement, and are plotted60S geographically60S in Figs. 3 and 60S 60S

90S 90S 90S 90S CCSM H 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E ble means are180W 150W 120W denoted90W 60W 30W 0 30E 60E 90E 120E by150E 180E square180W 150W 120W 90W 60W 30W brackets,0 30E 60E 90E 120E 150E 180E such as [LAT].

GISS - 4* CO2 anomaly

4, and latitudinally in Figs. 5 and 7. 90N

60N Eocene quantities are given a subscript e, and present/pre-

The SST plots have error bars which include30N the contri- 0 industrial (i.e. modern) quantities are given a subscript p. butions from the two sources of uncertainty30S we have con- 60S Model values are given a superscript m, and proxy or ob-

90S sidered, related toGISS calibration and temporal trends.180W 150W 120W 90W 60W 30W This0 30E 60E 90E 120E ap-150E 180E served data are given a superscript d. Because the modern proach to the data aims to include a wide range of potential observed data has global coverage (albeit interpolated, or uncertainties in order to highlight both the regions of poten- -20.0 -16.0 -12.0 -8.0 -4.0 0.0 assimilated4.0 8.0 with12.0 models16.0 in 20.0 some regions), but the Eocene tial model–data agreement, but more importantly whereTemperature there (Celsius) proxy data is sparse, the modern observed global or zonal appear to be genuine discrepancies that cannot realistically hT di T d be explained by the uncertainties in the proxy temperature means p and p are defined, but the Eocene equivalents are not. m m Fig. 4.estimations.Continental surface air temperature anomaly in the model simulations (LATe −GATp ), as a function of model and fractional CO2 d d increase from pre-industrial. Also shown for the proxies are LATe −GATp .

Clim. Past, 8, 1717–1736, 2012 www.clim-past.net/8/1717/2012/ 16 Lunt et al: EoMIP

×2 ×4 ×6 ×8 ×16

HadCM3L - 2* CO2 anomaly HadCM3L - 4* CO2 anomaly HadCM3L - 6* CO2 anomaly

90N 90N 90N

60N 60N 60N

30N 30N 30N

0 0 0

30S 30S 30S

60S 60S 60S

90S 90S 90S HadCM 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

ECHAM5 - 2* CO2 anomaly

90N

60N

30N

0

30S

60S

90S ECHAM 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

CCSM3_W - 4* CO2 anomaly CCSM3_W - 8* CO2 anomaly CCSM3_W - 16* CO2 anomaly

90N 90N 90N

60N 60N 60N

30N 30N 30N

0 0 0

30S 30S 30S

60S 60S 60S

90S 90S 90S CCSM W 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

CCSM_H - 2* CO2 anomaly CCSM_H - 4* CO2 anomaly CCSM_H - 8* CO2 anomaly CCSM_H - 16* CO2 anomaly

90N 90N 90N 90N

60N 60N 60N 60N

30N 30N 30N 30N

0 0 0 0

30S 30S 30S 30S

60S 60S 60S 60S

90S 90S 90S 90S CCSM H 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

GISS - 4* CO2 anomaly

90N

60N

30N

0

30S

60S

90S GISS 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

-20.0 -16.0 -12.0 -8.0 -4.0 0.0 4.0 8.0 12.0 16.0 20.0 Temperature (Celsius)

m m Fig. 3. SST anomaly in the model simulations (SSTe −SSTp ), as a function of model and fractional CO2 increase from pre-industrial. d d Also shown for the proxies are SSTe −SSTp .

D. J. Lunt et al.: EoMIP 1725

×2 ×4 ×6 ×8 ×16

HadCM3L - 2* CO2 anomaly HadCM3L - 4* CO2 anomaly HadCM3L - 6* CO2 anomaly

90N 90N 90N

60N 60N 60N

30N 30N 30N

0 0 0

30S 30S 30S

60S 60S 60S

90S 90S 90S HadCM 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

ECHAM5 - 2* CO2 anomaly

90N

60N

30N

0

30S

60S

90S ECHAM 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

CCSM3_W - 4* CO2 anomaly CCSM3_W - 8* CO2 anomaly CCSM3_W - 16* CO2 anomaly

90N 90N 90N

60N 60N 60N

30N 30N 30N

0 0 0

30S 30S 30S

60S 60S 60S

90S 90S 90S CCSM W 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

CCSM_H - 2* CO2 anomaly CCSM_H - 4* CO2 anomaly CCSM_H - 8* CO2 anomaly CCSM_H - 16* CO2 anomaly

90N 90N 90N 90N

60N 60N 60N 60N

30N 30N 30N 30N

0 0 0 0

30S 30S 30S 30S

60S 60S 60S 60S

90S 90S 90S 90S CCSM H 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

GISS - 4* CO2 anomaly

90N

60N

30N

0

30S

60S

90S GISS 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

-20.0 -16.0 -12.0 -8.0 -4.0 0.0 4.0 8.0 12.0 16.0 20.0 Temperature (Celsius)

m m Fig. 4. Continental surface air temperature anomaly in the model simulations (LATe − GATp ), as a function of model and fractional CO2 d d m m Fig. 4.increaseContinental from pre-industrial. surface air temperature Also shown anomaly for the proxies in the are model LAT simulationse − GATp. (LATe −GATp ), as a function of model and fractional CO2 d d increase from pre-industrial. Also shown for the proxies are LATe −GATp .

Table 2. Global mean temperatures and model mean error scores for temperatures (present; 1981–2010). For any given CO2 level, each simulation. Scores are calculated based on the SST (σsst) and there is a wide range of modelled Eocene global mean values; land surface air temperature (σ ) data. Definitions of the scores are ◦ m lat for example, at 560 ppmv, there is a 8.9 C range in hLATe i given in Eqs. (1) and (2). Rows in bold indicate the best (i.e. lowest ◦ m and a 3.2 C range in hSSTe i. This range is larger than σ ) CO2 level for each model. the range of simulated modern global means, which them- selves agree well with the observed modern global means. Model CO hSSTi hLATi hGSTi σ (◦C) σ (◦C) 2 sst lat The spread in Eocene results is due to (a) differences in HadCM 2× 21.45 11.71 18.54 8.8 15.5 the way the Eocene boundary conditions have been imple- × 4 24.19 16.20 21.95 6.0 11.4 mented in different models, and (b) different climate sen- 6× 26.25 19.80 24.56 3.9 7.7 sitivities in the different models. These differences are ex- ECHAM 2× 24.65 20.59 24.03 5.8 9.7 plored in Sect. 5. The clustering of the pre-industrial results × CCSM W 4 22.24 16.26 20.95 6.7 10.3 is likely a result of tuning of the pre-industrial simulations to 8× 24.45 19.57 23.59 4.0 7.2 16× 27.14 23.16 26.46 0.9 3.7 best match observations. For those models with more than CCSM H 2× 22.15 15.71 21.12 8.6 11.5 one Eocene simulation, the Eocene climate sensitivity (1 4× 23.94 18.41 23.17 6.6 8.5 hGATi per CO2 doubling) can also be seen to vary, both be- 8× 26.43 21.66 25.79 3.6 5.1 tween models and also within one model as a function of 16× 29.75 26.30 29.47 0.0 0.4 CO2. The variation of climate sensitivity between models is GISS 4× 26.43 21.97 23.25 3.8 6.9 well documented in the context of future climate simulations (e.g. IPCC, 2007). The increase in climate sensitivity with CO2 (for example in the CCSM H model) is due to the non- 4.1 Inter-model comparison linear behaviour of climate system feedbacks, for example, associated with water vapour (see Sect. 5); however, there Figure 2 shows the global annual mean sea surface tem- is also some non-linearity in the forcing itself as CO2 in- perature, hSSTi, and global annual mean near-surface land creases (Colman and McAveney, 2009). For HadCM, it is air temperature, hLATi, from all the GCM simulations in also related to a switch in ocean circulation which occurs be- the EoMIP ensemble, and for modern observations; the tween ×2 and ×4 CO2 and is associated with a non-linear in- Eocene values are also given in Table 2. The observed mod- crease in surface ocean temperature (Lunt et al., 2010). The ern datasets are HadISST for SSTs (pre-industrial; 1850– HadCM model also carried out an Eocene simulation with 1890) and NCEP (Kalnay et al., 1996) for near-surface air

www.clim-past.net/8/1717/2012/ Clim. Past, 8, 1717–1736, 2012 Lunt et al: EoMIP 17

1726 D. J. Lunt et al.: EoMIP

Proxy/model temperatures [degrees C] - HadCM3L Proxy/model temperatures [degrees C] - ECHAM5 Proxy/model temperatures [degrees C] - CCSM3_W 50 50 50 mean SST mean SST mean SST 2* CO2 2* CO2 4* CO2 4* CO2 1* CO2 8* CO2 6* CO2 16* CO2 40 1* CO2 40 40 1* CO2

30 30 30

20 20 20 temperature temperature temperature

10 10 10

0 0 0

-50 0 50 -50 0 50 -50 0 50 latitude latitude latitude (a) (b) (c)

Proxy/model temperatures [degrees C] - CCSM_H Proxy/model temperatures [degrees C] - GISS 50 50 mean SST mean SST 2* CO2 4* CO2 4* CO2 1* CO2 8* CO2 40 16* CO2 40 1* CO2

30 30

20 20 temperature temperature

10 10

0 0

-50 0 50 -50 0 50 latitude latitude (d) (e)

Fig. 5. Comparison of modelled SST with proxy-derived temperatures, SST vs. latitude. The simulations at ×1 CO2 are pre-industrial referenceFig. 5. Comparison simulations. of For modelled the model SST results, with proxy-derived the continous temperatures, lines represent SST the vs zonal. latitude. mean, The and simulationsthe open symbols at ×1 represent CO2 are pre-industrial the modelled temperaturereference simulations. at the same For location the model (longitude, results, latitude) the continous as the lines proxy repres data.ent For the the zonal proxy mean, data, and the the filled open symbols symbols represent represent the the mean modelled proxy temperature,temperature at and the the same error location bars represent (longitude,latitude) the range. The as smaller the proxy filled data. symbols For the are proxy EECO data, temperatures. the filled symbolsSee Sect. represent3 and the the Supplement mean proxy for moretemperature, details on and the the range error calculations. bars represent the range. The smaller filled symbols are EECO temperatures. See Section 3 and Supplementary Information for more details of the range calculations.

◦ ×1 CO2 (not shown). Comparison of that simulation with close to 0 C combined with reduced warming in the tropics its pre-industrial control shows that changing the non-CO2 due to a lack of snow and sea ice albedo feedbacks. How- boundary conditions to those of the Eocene (i.e. topographic, ever, other patterns are not consistent. For example, GISS bathymetric, vegetation, and solar constant changes) results at ×4 and HADCM at ×6 have similar values of hSSTi in a 1.8 ◦C increase in global mean surface air temperature, relative to their controls (8.6 and 9.0 ◦C, respectively), but ◦ by comparison with a 3.3 C increase for a CO2 doubling the warming in GISS is greatest in the northeast Pacific and from ×1 to ×2 under Eocene conditions. At a given CO2 the Southern Ocean, and the warming in HADCM is great- level, the CCSM W and CCSMProxy/modelH temperatures models [degrees give quite C] - HadCM3L differ- est inProxy/model the North temperatures Atlantic [degrees and C] - CCSM_H west of Australia. Similarly, 50 50 mean SST mean SST 2* CO2 2* CO2 ent global means. This difference4* CO2 in mean Eocene climate ECHAM4* CO2 at ×2 and CCSM H at ×4 have similar global mean 6* CO2 8* CO2 ◦ state between the two40 similar models is mostly due to dif- 40SST16* anomalies CO2 (7.2 and 7.6 C, respectively), but the greatest ferences in the assumed Eocene atmospheric aerosol load- Northern Hemisphere warming is in the Atlantic for ECHAM ing; CCSM W includes30 modern aerosols, whereas CCSM H 30and in the Pacific for CCSM H. The two CCSM models ex- includes no aerosol loading20 (see Sect. 2 and Table 1). Both 20hibit similar patterns of warming, correcting for their offset these models share thetemperature same pre-industrial simulation. For all temperature in absolute Eocene temperature – i.e. the patterns of warming models, the hLATi and10 hSSTi means share similar character- 10in CCSM H at ×8 are similar to those in CCSM W at ×16 istics, albeit with hSSTi varying over a smaller temperature (with anomalies of 10.1 and 10.8 ◦C, respectively). 0 0 range. Figure 4 shows the simulated annual mean LAT anomaly -50 0 50 -50 0 50 Figure 3 shows the simulated annual meanlatitude SST anomaly from each modellatitude and for the proxy reconstructions. The from each model and for the proxy reconstructions. A sim- anomaly is calculated relative to the pre-industrial (or mod- (a) (b) ple anomaly SSTe − SSTp would not be particularly infor- ern in the case of the proxies) global (land plus ocean) zonal mative because many regions would be undefined due to mean air temperature for each model, i.e. LATe − GATp. the difference in continental positions between the Eocene The global (as opposed to land-only) zonal mean is used for Fig. 6. As Figure 5a and d, but the HadCM and CCSM H modelled zonal mean represents the warm month mean SST, as opposed to annual − andmean. present. Instead, we show SSTe SSTp, which is only calculating the anomaly in order to avoid undefined points undefined over Eocene continental regions and latitudes at (for example in the latitudes of the Southern Ocean where which there is no ocean in the modern. The figures show there is no land in the modern). Similar to SST, there are that some features of temperature change are simulated con- some consistent features between models – greatest warming sistently across models, such as the greatest ocean warming is in the Antarctic (due to the lower topography via the lapse- occurring in the mid latitudes. This mid-latitude maximum rate effect and the change in albedo), and there is substantial is due to reduced SST warming in the high latitudes due to boreal polar amplification. Again, there are also differences the presence of seasonal sea ice anchoring the temperatures between models. For example, GISS at ×4 and ECHAM at

Clim. Past, 8, 1717–1736, 2012 www.clim-past.net/8/1717/2012/ Lunt et al: EoMIP 17

Proxy/model temperatures [degrees C] - HadCM3L Proxy/model temperatures [degrees C] - ECHAM5 Proxy/model temperatures [degrees C] - CCSM3_W 50 50 50 mean SST mean SST mean SST 2* CO2 2* CO2 4* CO2 4* CO2 1* CO2 8* CO2 6* CO2 16* CO2 40 1* CO2 40 40 1* CO2

30 30 30

20 20 20 temperature temperature temperature

10 10 10

0 0 0

-50 0 50 -50 0 50 -50 0 50 latitude latitude latitude (a) (b) (c)

Proxy/model temperatures [degrees C] - CCSM_H Proxy/model temperatures [degrees C] - GISS 50 50 mean SST mean SST 2* CO2 4* CO2 4* CO2 1* CO2 8* CO2 40 16* CO2 40 1* CO2

30 30

20 20 temperature temperature

10 10

0 0

-50 0 50 -50 0 50 latitude latitude (d) (e)

Fig. 5. Comparison of modelled SST with proxy-derived temperatures, SST vs. latitude. The simulations at ×1 CO2 are pre-industrial reference simulations. For the model results, the continous lines represent the zonal mean, and the open symbols represent the modelled temperature at the same location (longitude,latitude) as the proxy data. For the proxy data, the filled symbols represent the mean proxy temperature, and the error bars represent the range. The smaller filled symbols are EECO temperatures. See Section 3 and Supplementary Information for more details of the range calculations.

D. J. Lunt et al.: EoMIP 1727

Proxy/model temperatures [degrees C] - HadCM3L Proxy/model temperatures [degrees C] - CCSM_H 50 50 mean SST mean SST 2* CO2 2* CO2 4* CO2 4* CO2 6* CO2 8* CO2 40 40 16* CO2

30 30

20 20 temperature temperature

10 10

0 0

-50 0 50 -50 0 50 latitude latitude (a) (b)

Fig. 6. As Fig. 5a and d, but the HadCM and CCSM H modelled zonal mean represents the warm month mean SST as opposed to annual mean. Fig. 6. As Figure 5a and d, but the HadCM and CCSM H modelled zonal mean represents the warm month mean SST, as opposed to annual mean. ×2 have similar values of hLATi relative to their controls CCSM H models. In this case, the modelled warm month (8.5 and 7.3 ◦C, respectively), but GISS has a substantially mean temperature is within the uncertainty range of the Arc- greater warming over Southeast Asia. These differences can- tic TEX86 temperatures for both models. not be explained solely by differences in topography – the Figure 7 shows the terrestrial temperature model–data GISS and ECHAM models both use the Eocene topography comparison for each model. Those models which have been of Bice and Marotzke (2001). run at high CO2 (both CCSM models), show good agreement with the data across all latitudes. The other models do not 4.2 Model–data comparison simulate such high temperatures, but, as with SST, it does appear that if they had been run at higher CO2, the model– Figure 5 shows a zonal SST model–data comparison for data agreement would have been better. The HadCM model each model. The longitudinal locations of the SST data can appears to be somewhat of an outlier in the Northern Hemi- be seen in Fig. 3. Each model is capable of simulating sphere high latitudes, as it shows less polar amplification than Eocene SSTs which are within the uncertainty estimates of the other models (see Sect. 4.3), an effect also seen in SST. the majority of the data points. The data points which lie A quantitative indication of the model–data comparison furthest from the model simulations are the ACEX TEX86 for each simulation cannot currently be used to rank the 18 Arctic SST estimate (Sluijs et al., 2006), and the δ O and models themselves, because the actual CO2 forcing is not TEX86 estimates from the southwest Pacific (Bijl et al., well constrained by data. However, it could give an indication 2009). The Arctic temperature reconstructions have uncer- of the range of CO2 concentrations which are most consistent tainty estimates which mean that at high CO2, the CCSM H with the data. Given the sparseness of the SST and terrestrial (×8–16) and CCSM W(×16) model simulations are just data, any score should be treated with some caution. This is within agreement. At this CO2 level, these models are also confounded by the uneven spread of the data; for example, consistent with most of the tropical temperature estimates. there is a relatively high concentration of terrestrial data in From Fig. 2a, it is likely that other models could also obtain North America. There are also issues associated with the dif- similarly high Arctic temperatures if they were run at suf- ferent land–sea masks in the different models, which mean ficiently high CO2 or low aerosol forcing. Also, given that that the number of proxy data locations at which there are de- some of these models (e.g. HadCM) have a higher climate fined modelled values differs between the models. Therefore, sensitivity than CCSM H, this model–data consistency could we generate a simple mean error score for each simulation, be potentially obtained at a lower CO2 than in CCSM. σ, for both SST (σsst) and land air temperature (σlat) by av- TEX86 is a relatively new proxy, which, as discussed in eraging the error in temperature anomaly at the location of Sect. 3, is currently undergoing a process of rapid develop- each of N data points: ment. In this context, it has been suggested that the proxy 1 X m m d d could be recording the paleotemperature anomaly of the σsst = (SSTe − SSTp − SSTe + SSTp), (1) bloom season of the marine archaeota as opposed to a true N annual mean. If this is the case, then it is likely that a more 1 X m m d d σlat = (LATe − GATp − LATe + GATp), (2) appropriate comparison is with the modelled summer tem- N perature. This is illustrated in Fig. 6, for the HadCM and

www.clim-past.net/8/1717/2012/ Clim. Past, 8, 1717–1736, 2012 18 Lunt et al: EoMIP

1728 D. J. Lunt et al.: EoMIP

Proxy/model temperatures [degrees C] - HadCM3L Proxy/model temperatures [degrees C] - ECHAM5 Proxy/model temperatures [degrees C] - CCSM3_W proxy surface temperature proxy surface temperature proxy surface temperature 2* CO2 2* CO2 4* CO2 40 4* CO2 40 1* CO2 40 8* CO2 6* CO2 16* CO2 1* CO2 1* CO2

20 20 20 temperature temperature temperature

0 0 0

-20 -20 -20

-50 0 50 -50 0 50 -50 0 50 latitude latitude latitude (a) (b) (c)

Proxy/model temperatures [degrees C] - CCSM_H Proxy/model temperatures [degrees C] - GISS proxy surface temperature proxy surface temperature 2* CO2 4* CO2 40 4* CO2 40 1* CO2 8* CO2 16* CO2 1* CO2

20 20 temperature temperature

0 0

-20 -20

-50 0 50 -50 0 50 latitude latitude (d) (e)

Fig. 7. Comparison of modelled SAT with proxy-derived temperatures, SAT vs. latitude. The simulations at ×1 CO2 are pre-industrial refer- ence simulations. For the model results, the continous lines represent the zonal mean, and the symbols represent the modelled temperature at the same location (longitude, latitude) as the proxy data. For the proxy data, the symbols represent the proxy temperature, and the error bars Fig. 7. Comparison of modelled SAT with proxy-derived temperatures, SAT vs. latitude. The simulations at ×1 CO2 are pre-industrial ref- represent the range, as given by Huber and Caballero (2011). erence simulations. For the model results, the continous lines represent the zonal mean, and the symbols represent the modelled temperature at the same location (longitude,latitude) as the proxy data. For the proxy data, the symbols represent the proxy temperature, and the error bars represent the range,but proceed as given with by caution, Huber being and mindful Caballero that there (2011) is a con-. simulation) to provide a first-order estimate of the CO2 level, siderable uncertainty in the score itself. Values of σ for each for each model, which could give the best agreement with model simulation are given in Table 2. For each model, the the proxy estimates. For HadCM, CCSM H, and CCSM W, best results are obtained for the highest CO2 level which was using σsst this is 2600 ppmv, 4300 ppmv, and 4900 ppmv, re- simulated (a result which also applies if an RMS score is spectively, and using σlat this is 2800 ppmv, 4500 ppmv, and used in place of a mean error score). The CCSM H model 6300 ppmv, respectively. These estimates come with many at 16× CO2 has the best (i.e. lowest absolute) values of σ . caveats, not least that the uneven and sparse data spread However, as noted before, it appears that other models would means that the absolute minimum mean error, σ, is not nec- also obtain good σ scores if they had been run at sufficiently essarily a good indicator of the correct global mean tempera- high CO2. A “best-case” multi-model ensemble can be cre- ture. However, they do indicate the magnitude of the range of ated by averaging the simulations from each model which CO2 values that could be considered consistent with model have the lowest values of σ (it turns out that those mod- results. These values are significantly higher than those pre- Proxy/model temperatures [degrees C] Proxy/model temperatures [degrees C] els with50 the best σlat also have the best σsst). These are the sented50 for this time period in the compilation of Beerling and models highlighted in bold in Table 2. The model–data com- Royer (2011).

parison40 for this multi-model ensemble is shown in Figs. 8 40 and 9. The 2 standard-deviation width of the “best-case” 4.3 Meridional gradients and polar amplification ensemble overlaps the uncertainty estimates of every terres- trial30 and ocean proxy data point. However, the high latitude The30 changes in meridional temperature gradient are sum- southwest Pacific SST estimates are right at the boundary of marised in Fig. 10, which shows the surface temperature dif- consistency. The terrestrial data shows very good agreement 20 | | ◦

temperature ferencetemperature between the low latitudes ( φ < 30 ) and the high with the model ensemble, and both data and models show a 20 latitudes (|φ| > 60◦) as a function of global mean temper- similar degree of polar amplification (see Sect. 4.3). ature, and how this is partitioned between land and ocean By10 regressing the CO levels and σ values in Table 2, it 2 warming (Fig. 10b, φ is latitude in degrees). All the Eocene is possible (for those models with more than one Eocene 10 simulations have a reduced meridional surface temperature 0

0 -50 0 50 -50 0 50 Clim. Past, 8, 1717–1736, 2012latitude www.clim-past.net/8/1717/2012/latitude

(a) (b)

Fig. 8. Zonal ensemble mean model (middle thick black line), and data, presented as an anomaly relative to present/pre-industrial. Outer thick black lines indicate ±2 standard deviations in the models. Coloured lines represent each individual model simulation in the ensemble, with the colour indicating CO2 level as in Figures 5 and 7. (a) [SSTe −SSTp]. (b) [LATe −GATp]. For this Figure, the ensemble consists of the best simulation from each model, as highlighted in bold in Table 2. Descriptions of the proxy error bars are given in the captions to Figures 5 and 7. 18 Lunt et al: EoMIP

Proxy/model temperatures [degrees C] - HadCM3L Proxy/model temperatures [degrees C] - ECHAM5 Proxy/model temperatures [degrees C] - CCSM3_W proxy surface temperature proxy surface temperature proxy surface temperature 2* CO2 2* CO2 4* CO2 40 4* CO2 40 1* CO2 40 8* CO2 6* CO2 16* CO2 1* CO2 1* CO2

20 20 20 temperature temperature temperature

0 0 0

-20 -20 -20

-50 0 50 -50 0 50 -50 0 50 latitude latitude latitude (a) (b) (c)

Proxy/model temperatures [degrees C] - CCSM_H Proxy/model temperatures [degrees C] - GISS proxy surface temperature proxy surface temperature 2* CO2 4* CO2 40 4* CO2 40 1* CO2 8* CO2 16* CO2 1* CO2

20 20 temperature temperature

0 0

-20 -20

-50 0 50 -50 0 50 latitude latitude (d) (e)

Fig. 7. Comparison of modelled SAT with proxy-derived temperatures, SAT vs. latitude. The simulations at ×1 CO2 are pre-industrial ref- erence simulations. For the model results, the continous lines represent the zonal mean, and the symbols represent the modelled temperature at the same location (longitude,latitude) as the proxy data. For the proxy data, the symbols represent the proxy temperature, and the error bars represent the range, as given by Huber and Caballero (2011).

D. J. Lunt et al.: EoMIP 1729

Proxy/model temperatures [degrees C] Proxy/model temperatures [degrees C] 50 50

40 40

30 30

20 temperature temperature 20

10

10

0

0 -50 0 50 -50 0 50 Lunt et al: EoMIP latitude latitude 19 (a) (b)

Fig. 8. Zonal ensemble mean model (middle thick black line), and data, presented as an anomaly relative to present/pre-industrial. Outer thick black lines indicate ±2 standard deviations in the models. Coloured lines represent each individual model simulation in the ensemble, Fig. 8. Zonal ensemble mean model (middle thick black line), and data, presented as an anomaly relative to present/pre-industrial. Outer with the colour indicating CO2 level as in Figs. 5 and 7. (a) [SSTe −SSTp]. (b) [LATe −GATp]. For this figure, the ensemble consists of the thick blackbest lines simulation indicate from±2 each standard model, deviations as highlighted in thein bold models. in Table Coloured2. Descriptions lines represent of the proxy each error indiv barsidual are given model in the simulation captions to in Figs. the5 ensemble, with the colourand 7. indicating CO2 level as in Figures 5 and 7. (a) [SSTe −SSTp]. (b) [LATe −GATp]. For this Figure, the ensemble consists of the best simulation from each model, as highlighted in bold in Table 2. Descriptions of the proxy error bars are given in the captions to

Eocene_ensemble Eocene_ensemble Figures 5 and 7. 90N 90N

60N 60N

30N 30N

0 0

30S 30S

60S 60S

90S 90S 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

-20.0 -16.0 -12.0 -8.0 -4.0 0.0 4.0 8.0 12.0 16.0 20.0 -20.0 -16.0 -12.0 -8.0 -4.0 0.0 4.0 8.0 12.0 16.0 20.0 Temperature (Celsius) Temperature (Celsius)

(a) (b)

Fig. 9. Ensemble mean modelled Eocene warming, presented as an anomaly relative to present/pre-industrial. (a) [SSTe−SSTp]. (b) [LATe− GATp]. The ensemble consists of the best simulation from each model, as highlighted in bold in Table 2. Fig. 9. Ensemble mean modelled Eocene warming, presented as an anomaly relative to present/pre-industrial. (a) [SSTe − SSTp]. (b) [LATe −GATp]. The ensmble consists of the best simulation from each model, as highlighted in bold in Table 2. gradient compared with the pre-industrial, and the gradi- models, the meridional temperature gradient is reduced by a ent reduces further as CO2 increases, i.e. polar amplifica- similar amount over land and ocean as a function of CO2, tion increases (Fig. 10a). However, there is a high degree with some indication, at maximum (×16) CO2, that the SST of inter-model variability in the absolute Eocene gradient, gradient starts reducing more over ocean than over land. This with HadCM appearing to be an outlier with a relatively high implies that when considering changes relative to the mod- Eocene gradient. There is some indication that the models are ern, it is possible to have substantially different temperature trending towards a minimum gradient of about 20 ◦C. This, changes over land compared with over ocean at the same lat- along with our energy flux analysis (see Sect. 5), supports itude. This is also clear from comparing Fig. 3 with Fig. 4, previous work (Huber et al., 2003) that implied that merid- and shows the importance of differentiating terrestrial and ional temperature gradients of the order 20 ◦C were physi- oceanic signals when considering the consistency between cally realistic, even without large changes to meridional heat different proxy data, and between data and models. transport. Compared with pre-industrial, the meridional sur- face temperature gradient reduces more on land than over 50 45 ocean (Fig. 10b). For HadCM, this applies also to the Eocene simulations as CO2 increases. However, for the two CCSM 40 40

www.clim-past.net/8/1717/2012/35 Clim. Past, 8, 1717–1736, 2012

30 30

20 25 meridonal temperature gradient meridional temperature gradient [sst]

20 10 280 285 290 295 300 305 20 25 30 35 40 45 global average surface temperature meridional temperature gradient [land]

(a) (b)

Fig. 10. (a) Meridional surface temperature gradient GST|φ|>60 −GST|φ|<30, where |φ| is the absolute value of the latitude in degrees, as a function of global mean surface temperature, hGST i for all the simulations presented in this paper. (b) Meridional surface temperature gradient over land / ocean, SST|φ|>60 −SST|φ|<30 vs. LAT|φ|>60 −LAT|φ|<30. Symbols and colours correspond to those in Figure 2. Lunt et al: EoMIP 19

Eocene_ensemble Eocene_ensemble

90N 90N

60N 60N

30N 30N

0 0

30S 30S

60S 60S

90S 90S 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E 180W 150W 120W 90W 60W 30W 0 30E 60E 90E 120E 150E 180E

-20.0 -16.0 -12.0 -8.0 -4.0 0.0 4.0 8.0 12.0 16.0 20.0 -20.0 -16.0 -12.0 -8.0 -4.0 0.0 4.0 8.0 12.0 16.0 20.0 Temperature (Celsius) Temperature (Celsius)

(a) (b)

Fig. 9. Ensemble mean modelled Eocene warming, presented as an anomaly relative to present/pre-industrial. (a) [SSTe − SSTp]. (b) [LATe −GATp]. The ensmble consists of the best simulation from each model, as highlighted in bold in Table 2.

1730 D. J. Lunt et al.: EoMIP

50 45

40 40

35

30 30

20 25 meridonal temperature gradient meridional temperature gradient [sst]

20 10 280 285 290 295 300 305 20 25 30 35 40 45 global average surface temperature meridional temperature gradient [land]

(a) (b)

Fig. 10. (a) Meridional surface temperature gradient GST|φ|>60 − GST|φ|<30, where |φ| is the absolute value of the latitude in degrees, as a function of global mean surface temperature, hGSTi, for all the simulations presented in this paper. (b) Meridional surface temperature Fig. 10. (a) Meridional surface temperature gradient GST φ >60 −GST φ <30, where |φ| is the absolute value of the latitude in degrees, as gradient over land/ocean, SST − SST vs. LAT| | − LAT| | . Symbols and colours correspond to those in Fig. 2. a function of global mean surface|φ temperature,|>60 |φ|h the60 simulations|φ|<30 presented in this paper. (b) Meridional surface temperature gradient over land / ocean, SST|φ|>60 −SST|φ|<30 vs. LAT|φ|>60 −LAT|φ|<30. Symbols and colours correspond to those in Figure 2. ↓ 5 Reasons for inter-model variability: an energy flux where SWt is the incoming solar radiation at the top of the analysis atmosphere, α is the planetary albedo, H is the net merid- ional heat transport convergence,  is the atmospheric emis- It is interesting up to a point to simply intercompare model sivity, σ is the Stephan–Boltzmann constant, and τ is the sur- results, and to compare with data, but also of interest is to face temperature, to be diagnosed by the EBM. All variables know why different models behave differently. Given the are functions of latitude apart from σ. complexity of climate models, this can be problematic, and The planetary albedo is given by traditionally, groups such as PMIP have not often diagnosed ↑ in detail the differences. Here, we attempt to diagnose some SW = t aspects of the differences between the model results, building α ↓ , (4) SW on a 1-D energy-balance approach as outlined by Heinemann t et al. (2009). Here, the causes of the zonal mean temperature and the atmospheric emissivity is given by response of a model are diagnosed from the top of the at- ↑ mosphere and surface radiative fluxes, including their clear- LW = t sky values, assuming simple energy balance. Any difference  ↑ , (5) LW between the meridional temperature profile in the GCM and s that estimated from the energy-balance approach is attributed ↑ ↓ where SWt and SWt are the outgoing and incoming top of to meridional heat transport. As such, the change in merid- ↑ the atmosphere shortwave radiation, respectively, and LWt ional temperature profile between two simulations (such as a ↑ pre-industrial control and an Eocene simulation) can be at- and LWs are the upwelling top of the atmosphere and surface tributed to a combination of (1) changes in emissivity due to longwave radiation, respectively. Given that the surface emits changes in clouds, (2) changes in emissivity due to changes long wave radiation according to in the greenhouse effect (i.e. CO and water vapour concen- 2 LW↑ = σ τ 4, (6) tration changes, and lapse-rate effects), (3) changes in albedo s due to changes in clouds, (4) changes in albedo due to Earth it follows that the meridional heat transport convergence, H, surface and atmospheric aerosol changes, and (5) changes in is given by meridional heat transport. net net Following Heinemann et al. (2009), the 1-D energy bal- H = −(SWt + LWt ), (7) ance model (EBM) is formulated by equating the incoming solar radiation with outgoing long wave radiation, with any where the superscript net represents net flux (positive down- local imbalance attributed to local meridional heat transport: wards). Equation (7) reflects the necessity that, in equilib- rium, any net downward shortwave plus longwave heat flux ↓ 4 has to be compensated by a negative meridional heat flux SWt (1 − α) + H = σ τ , (3) convergence (note that there is a typo in the equivalent equa- tion of Heinemann et al. (2009), the minus sign in their Eq. 4

Clim. Past, 8, 1717–1736, 2012 www.clim-past.net/8/1717/2012/ D. J. Lunt et al.: EoMIP 1731 is erroneous). All the radiative fluxes are output directly from by non-cloud albedo changes, with a significant contribution the GCMs, and used as input into the energy balance model. also from emissivity changes. In both these models, short- From Eq. (3), it follows that wave cloud albedo changes act to reduce the polar ampli- fication in both hemispheres. The greater global tempera- 1 ↓ 0.25 ture change in ECHAM compared with CCSM H is due to τ = ( (SWt (1 − α) + H) ) ≡ E(,α,H ). (8) σ the greater change in greenhouse effect. However, the en- The difference in temperature between two simulations, ergy balance analysis does not allow us to diagnose if this 1T = τ − τ 0, is given by E(,α,H ) − E(0,α0,H 0), where is due to a greater radiative forcing given the same CO2 in- the prime, 0, represents values in the second simulation. In or- crease, or due to greater water vapour feedbacks or lapse- der to diagnose the reasons for the temperature differences in rate changes in ECHAM. HadCM exhibits quite different be- two simulations, we consider changes to the diagnosed emis- haviour. In the Southern Hemisphere, the zonal mean tem- sivity, planetary albedo, and heat transport, and write perature increase is due predominantly to non-cloud albedo changes, and is reduced relative to the other two models. 0 1Temm = E(,α,H ) − E( ,α,H ) (9) In the Northern Hemisphere, the increase in temperature is 0 much reduced relative to the other two models due to a lack 1Talb = E(,α,H ) − E(,α ,H) (10) of non-cloud albedo feedbacks and changes in emissivity. 1T = E(,α,H ) − E(,α,H 0), (11) tran Abbot and Tziperman (2008) suggested that the lack of sea where 1Temm, 1Talb, and 1Ttran are the components of ice in the Arctic can lead to stronger convection over the 1T due to emissivity, planetary albedo, and heat transport relatively warm Arctic sea surface during winter, leading to changes, respectively. Because the changes in emissivity, more convective clouds and increased water vapour concen- albedo, and heat transport are relatively small compared to trations, and thereby causing polar amplification via both their magnitude, albedo and emissivity effects. The largely decreased (ver- sus unchanged) surface albedo in northern high latitudes in 1T ' 1Temm + 1Talb + 1Ttran. (12) CCSM H and ECHAM (versus HadCM), increased (versus virtually unchanged) longwave cloud radiative forcing, and We further partition the 1Temm and 1Talb terms by consid- reduced (versus hardly changed) clear-sky emissivity indi- ering the clear-sky radiative fluxes, also output directly from cate that this sea ice/convection feedback is active for ×1 to the GCMs. Using cs as a subscript to denote clear-sky fluxes, ×2 in CCSM H and ECHAM, but absent in HadCM. The re- we can estimate the contribution due to the greenhouse effect duced strength of this feedback in HadCM may be related to (CO2 and water vapour and lapse rate) changes, 1Tgg, and the relatively strong Eocene seasonality in HadCM compared the contribution due to surface albedo and aerosol changes, with the other models (as can be seen by comparing Fig. 6 1Tsalb: with Fig. 5), which suppresses Arctic convection in HadCM in winter. Changes in heat transport are playing a relatively 1T = E( ,α ,H ) − E(0 ,α ,H ) (13) gg cs cs cs cs cs cs minor role in determining the latitudinal temperature profile = − 0 1Tsalb E(cs,αcs,Hcs) E(cs,αcs,Hcs), (14) in ECHAM, CCSM H, and HadCM, which supports previ- ous findings using ECHAM alone (Heinemann et al., 2009). because the emissivity change in the clear-sky case is solely Figure 11d–g shows the models which simulate a transi- due to greenhouse effect changes, and the albedo change tion from pre-industrial to Eocene at ×4 CO . For HadCM in the clear-sky case is mainly due to surface albedo and 2 and CCSM H, the results are very similar to ×2 CO , but aerosols. Considering the remaining temperature difference 2 with greater magnitude; for both models each component as due to clouds, we can then write contributes the same fraction to the total warming under ×2 1Tlwc = 1Temm − 1Tgg (15) as to under ×4, to within ∼ 10 %. CCSM W is very simi- lar to CCSM H, except that it has reduced warming due to 1Tswc = 1Talb − 1Tsalb, (16) decreased change in non-cloud albedo. This is most likely a where 1Tlwc and 1Tswc are the components of 1T due to direct result of the different aerosol fields applied in the these long-wave cloud changes and short-wave cloud changes, re- two models for the Eocene (see Table 1). The model which spectively. In this way, a temperature difference between two exhibits the greatest warming is the GISS model. This high simulations can be partitioned into 5 components, given by sensitivity relative to the other models is due to greater green- Eqs. (9)–(11) and (13)–(16). house gas effect changes and greater cloud albedo feedbacks. Figure 11 shows the results from this energy balance anal- The warming over Antarctica is particularly large in the GISS ysis for a number of pairs of simulations. Figure 11a–c shows model, and is due to a greater local change in non-cloud the models which simulate a transition from pre-industrial albedo. However, the GISS model also has strong negative to Eocene at ×2 CO2. ECHAM and CCSM H show similar cloud forcing at high latitudes in both hemispheres. results in terms of the reasons for this change. They show Figure 11h–i shows the models which simulate a transi- a high latitude warming in both hemispheres caused mainly tion from ×2 to ×4 CO2 under Eocene conditions. HadCM www.clim-past.net/8/1717/2012/ Clim. Past, 8, 1717–1736, 2012 20 Lunt et al: EoMIP

1732 D. J. Lunt et al.: EoMIP

Surface temperature differences between 1* CO2 and 2* CO2 - HadCM3L Surface temperature differences between 1* CO2 and 2* CO2 - ECHAM5 Surface temperature differences between 1* CO2 and 2* CO2 - CCSM_H GCM: 5.2 GCM: 9.3 GCM: 8.1 40 EBM: 5.1 40 EBM: 9.2 40 EBM: 8.1 emissivity due to clouds: 0.1 1.3% emissivity due to clouds: -0.0 -0.4% emissivity due to clouds: 0.3 3.2% emissivity due to greenhouse gases: 3.3 63.6% emissivity due to greenhouse gases: 6.1 66.7% emissivity due to greenhouse gases: 4.9 60.8% albedo due to clouds: 0.3 6.0% albedo due to clouds: -1.0 -11.2% albedo due to clouds: -1.1 -13.7% 30 30 30 albedo due to planetary surface: 1.7 33.4% albedo due to planetary surface: 3.8 41.1% albedo due to planetary surface: 4.1 50.9% due to heat transport -0.0 due to heat transport 0.2 due to heat transport -0.2

20 20 20

10 10 10

0 0 0 surface temperature difference surface temperature difference surface temperature difference

-10 -10 -10

-50 0 50 -50 0 50 -50 0 50 latitude latitude latitude (a) (b) (c)

Surface temperature differences between 1* CO2 and 4* CO2 - HadCM3L Surface temperature differences between 1* CO2 and 4* CO2 - GISS Surface temperature differences between 1* CO2 and 4* CO2 - CCSM_W Surface temperature differences between 1* CO2 and 4* CO2 - CCSM_H GCM: 8.6 GCM: 11.0 GCM: 7.9 GCM: 10.1 40 EBM: 8.5 40 EBM: 11.0 40 EBM: 7.9 40 EBM: 10.1 emissivity due to clouds: 0.1 0.9% emissivity due to clouds: -0.0 -0.4% emissivity due to clouds: 0.2 3.0% emissivity due to clouds: 0.2 2.3% emissivity due to greenhouse gases: 5.9 69.1% emissivity due to greenhouse gases: 8.3 75.5% emissivity due to greenhouse gases: 6.0 75.5% emissivity due to greenhouse gases: 7.1 70.1% albedo due to clouds: 0.8 9.1% albedo due to clouds: 2.4 21.9% albedo due to clouds: 0.1 1.2% albedo due to clouds: -1.4 -14.0% 30 30 30 30 albedo due to planetary surface: 2.1 24.1% albedo due to planetary surface: 0.1 1.1% albedo due to planetary surface: 1.8 23.3% albedo due to planetary surface: 4.3 42.2% due to heat transport -0.1 due to heat transport 0.3 due to heat transport -0.2 due to heat transport -0.3

20 20 20 20

10 10 10 10

0 0 0 0 surface temperature difference surface temperature difference surface temperature difference surface temperature difference

-10 -10 -10 -10

-50 0 50 -50 0 50 -50 0 50 -50 0 50 latitude latitude latitude latitude (d) (e) (f) (g)

Surface temperature differences between 2* CO2 and 4* CO2 - HadCM3L Surface temperature differences between 2* CO2 and 4* CO2 - CCSM_H 15 GCM: 3.4 15 GCM: 2.1 EBM: 3.4 EBM: 2.1 emissivity due to clouds: 0.0 0.3% emissivity due to clouds: -0.0 -1.5% emissivity due to greenhouse gases: 2.6 78.0% emissivity due to greenhouse gases: 2.2 108.1% albedo due to clouds: 0.5 13.9% albedo due to clouds: -0.3 -14.9% 10 10 albedo due to planetary surface: 0.3 10.1% albedo due to planetary surface: 0.2 8.5% due to heat transport -0.1 due to heat transport -0.0

5 5

0 0 surface temperature difference surface temperature difference

-5 -5 -50 0 50 -50 0 50 latitude latitude (h) (i)

Surface temperature differences between 4* CO2 and 8* CO2 - CCSM_W Surface temperature differences between 4* CO2 and 8* CO2 - CCSM_H 15 GCM: 2.6 15 GCM: 2.6 EBM: 2.6 EBM: 2.6 emissivity due to clouds: -0.0 -1.6% emissivity due to clouds: -0.0 -1.9% emissivity due to greenhouse gases: 2.7 101.6% emissivity due to greenhouse gases: 2.8 107.8% albedo due to clouds: -0.2 -9.1% albedo due to clouds: -0.3 -11.4% 10 10 albedo due to planetary surface: 0.3 9.7% albedo due to planetary surface: 0.1 5.1% due to heat transport -0.0 due to heat transport 0.0

5 5

0 0 surface temperature difference surface temperature difference

-5 -5 -50 0 50 -50 0 50 latitude latitude (j) (k)

Fig. 11. The zonal mean surface temperature change under a range of CO2 transitions, and energy balance analysis of the reasons for the changes. (a–c) ×1 to ×2 CO2, (d–g) ×1 to ×4 CO2, (h–i) ×2 to ×4 CO2, (j–k) ×4 to ×8 CO2. The simulations at ×1 CO2 are pre-industrial reference simulations. Note the difference in vertical scale in panels (a–g) compared with (h–k). The dotted lines in the plots show the sum Fig. 11. The zonal-mean surface temperature change under a range of CO2 transitions, and energy balance analysis of the reasons for the of the various components, which in each case should be very close to the GCM line (i.e. the “actual” temperature change from the model) changes. (a-c) ×1 toand×2 the CO EBM2, line (d-g) (i.e.×1τ1as to calculated×4 CO from2, (h-i) Eq. 8 for×2 the to two× climate4 CO states).2, (j-k) ×4 to ×8 CO2. The simulations at ×1 CO2 are pre-industrial reference simulations. Note the difference in vertical scale in panels (a-g) compared with (h-k). The dotted lines in the plots show the sum of the various components, which in each case should be very close to the GCM line (i.e. the ‘actual’ temperature change from the model) has a greater climate sensitivity that CCSM H, and this is Figure 11j–k shows the models which simulate a transi- and the EBM line (i.e.due∆ toτ greateras caluclated changes in from greenhouse Equation gas emissivity, 8 for the and two a climatetion from states).×4 to ×8 CO2 under Eocene conditions. Similar positive as opposed to negative cloud albedo feedback. The to the transition from ×2 to ×4, the polar amplification is relative lack of polar amplification in both models compared relatively small. The warming is due almost entirely to the to the results discussed above is due to the lack of Antarc- changes in emissivity (direct CO2 forcing and water vapour tic ice sheet in the Eocene. The small amount of polar am- feedbacks and lapse-rate changes), and unsurprisingly has plification in HadCM is due to changes in heat transport; in a similar latitudinal distribution in the two models. How- CCSM it is due to non-cloud albedo changes in the Northern ever, in the Northern Hemisphere high latitudes the CCSM H Hemisphere. model shows strong opposing effects of cloud and surface changes, which are not present in CCSM W. This is most Albedo vs latitude - HadCM3L Albedo vs latitude - ECHAM5 Albedo vs latitude - CCSM_H 0.8 0.8 0.8 2* CO2 2* CO2 2* CO2 1* CO2 1* CO2 1* CO2

Clim.0.6 Past, 8, 1717–1736, 20120.6 0.6 www.clim-past.net/8/1717/2012/

0.4 0.4 0.4 albedo albedo albedo

0.2 0.2 0.2

0.0 0.0 0.0 -50 0 50 -50 0 50 -50 0 50 latitude latitude latitude (a) (b) (c)

Fig. 12. Zonal mean albedo in the ×1 and ×2 CO2 simulations using the (a) HadCM, (b) ECHAM, and (c) CCSM H models. The simulations at ×1 CO2 are pre-industrial reference simulations. 20 Lunt et al: EoMIP

Surface temperature differences between 1* CO2 and 2* CO2 - HadCM3L Surface temperature differences between 1* CO2 and 2* CO2 - ECHAM5 Surface temperature differences between 1* CO2 and 2* CO2 - CCSM_H GCM: 5.2 GCM: 9.3 GCM: 8.1 40 EBM: 5.1 40 EBM: 9.2 40 EBM: 8.1 emissivity due to clouds: 0.1 1.3% emissivity due to clouds: -0.0 -0.4% emissivity due to clouds: 0.3 3.2% emissivity due to greenhouse gases: 3.3 63.6% emissivity due to greenhouse gases: 6.1 66.7% emissivity due to greenhouse gases: 4.9 60.8% albedo due to clouds: 0.3 6.0% albedo due to clouds: -1.0 -11.2% albedo due to clouds: -1.1 -13.7% 30 30 30 albedo due to planetary surface: 1.7 33.4% albedo due to planetary surface: 3.8 41.1% albedo due to planetary surface: 4.1 50.9% due to heat transport -0.0 due to heat transport 0.2 due to heat transport -0.2

20 20 20

10 10 10

0 0 0 surface temperature difference surface temperature difference surface temperature difference

-10 -10 -10

-50 0 50 -50 0 50 -50 0 50 latitude latitude latitude (a) (b) (c)

Surface temperature differences between 1* CO2 and 4* CO2 - HadCM3L Surface temperature differences between 1* CO2 and 4* CO2 - GISS Surface temperature differences between 1* CO2 and 4* CO2 - CCSM_W Surface temperature differences between 1* CO2 and 4* CO2 - CCSM_H GCM: 8.6 GCM: 11.0 GCM: 7.9 GCM: 10.1 40 EBM: 8.5 40 EBM: 11.0 40 EBM: 7.9 40 EBM: 10.1 emissivity due to clouds: 0.1 0.9% emissivity due to clouds: -0.0 -0.4% emissivity due to clouds: 0.2 3.0% emissivity due to clouds: 0.2 2.3% emissivity due to greenhouse gases: 5.9 69.1% emissivity due to greenhouse gases: 8.3 75.5% emissivity due to greenhouse gases: 6.0 75.5% emissivity due to greenhouse gases: 7.1 70.1% albedo due to clouds: 0.8 9.1% albedo due to clouds: 2.4 21.9% albedo due to clouds: 0.1 1.2% albedo due to clouds: -1.4 -14.0% 30 30 30 30 albedo due to planetary surface: 2.1 24.1% albedo due to planetary surface: 0.1 1.1% albedo due to planetary surface: 1.8 23.3% albedo due to planetary surface: 4.3 42.2% due to heat transport -0.1 due to heat transport 0.3 due to heat transport -0.2 due to heat transport -0.3

20 20 20 20

10 10 10 10

0 0 0 0 surface temperature difference surface temperature difference surface temperature difference surface temperature difference

-10 -10 -10 -10

-50 0 50 -50 0 50 -50 0 50 -50 0 50 latitude latitude latitude latitude (d) (e) (f) (g)

Surface temperature differences between 2* CO2 and 4* CO2 - HadCM3L Surface temperature differences between 2* CO2 and 4* CO2 - CCSM_H 15 GCM: 3.4 15 GCM: 2.1 EBM: 3.4 EBM: 2.1 emissivity due to clouds: 0.0 0.3% emissivity due to clouds: -0.0 -1.5% emissivity due to greenhouse gases: 2.6 78.0% emissivity due to greenhouse gases: 2.2 108.1% albedo due to clouds: 0.5 13.9% albedo due to clouds: -0.3 -14.9% 10 10 albedo due to planetary surface: 0.3 10.1% albedo due to planetary surface: 0.2 8.5% due to heat transport -0.1 due to heat transport -0.0

5 5

0 0 surface temperature difference surface temperature difference

-5 -5 -50 0 50 -50 0 50 latitude latitude (h) (i)

Surface temperature differences between 4* CO2 and 8* CO2 - CCSM_W Surface temperature differences between 4* CO2 and 8* CO2 - CCSM_H 15 GCM: 2.6 15 GCM: 2.6 EBM: 2.6 EBM: 2.6 emissivity due to clouds: -0.0 -1.6% emissivity due to clouds: -0.0 -1.9% emissivity due to greenhouse gases: 2.7 101.6% emissivity due to greenhouse gases: 2.8 107.8% albedo due to clouds: -0.2 -9.1% albedo due to clouds: -0.3 -11.4% 10 10 albedo due to planetary surface: 0.3 9.7% albedo due to planetary surface: 0.1 5.1% due to heat transport -0.0 due to heat transport 0.0

5 5

0 0 surface temperature difference surface temperature difference

-5 -5 -50 0 50 -50 0 50 latitude latitude (j) (k)

Fig. 11. The zonal-mean surface temperature change under a range of CO2 transitions, and energy balance analysis of the reasons for the changes. (a-c) ×1 to ×2 CO2, (d-g) ×1 to ×4 CO2, (h-i) ×2 to ×4 CO2, (j-k) ×4 to ×8 CO2. The simulations at ×1 CO2 are pre-industrial reference simulations. Note the difference in vertical scale in panels (a-g) compared with (h-k). The dotted lines in the plots show the sum of the various components, which in each case should be very close to the GCM line (i.e. the ‘actual’ temperature change from the model) and the EBM line (i.e. ∆τ as caluclated from Equation 8 for the two climate states).

D. J. Lunt et al.: EoMIP 1733

Albedo vs latitude - HadCM3L Albedo vs latitude - ECHAM5 Albedo vs latitude - CCSM_H 0.8 0.8 0.8 2* CO2 2* CO2 2* CO2 1* CO2 1* CO2 1* CO2

0.6 0.6 0.6

0.4 0.4 0.4 albedo albedo albedo

0.2 0.2 0.2

0.0 0.0 0.0 -50 0 50 -50 0 50 -50 0 50 latitude latitude latitude (a) (b) (c)

Fig. 12. Zonal mean albedo in the ×1 and ×2 CO2 simulations using the (a) HadCM, (b) ECHAM, and (c) CCSM H models. The simulations at ×1 CO2 are pre-industrial reference simulations. Fig. 12. Zonal mean albedo in the ×1 and ×2 CO2 simulations using the (a) HadCM, (b) ECHAM, and (c) CCSM H models. The simulations at ×1 CO2 are pre-industrial reference simulations. likely due to the remnants of Arctic sea ice in CCSM W the model–data agreement improves further. We have inter- at ×8 CO2 which are not present in the warmer CCSM H rogated the reasons for the differences between the models, model. Comparison of Fig. 11k with Fig. 11i shows that the and found differences in climate sensitivity to be due primar- increase in climate sensitivity in CCSM H as a function of ily to a combination of greenhouse effect and surface albedo background CO2 is due almost entirely to increased non- feedbacks, rather than differences in heat transport or cloud cloud emissivity changes; the framework does not allow us feedbacks. to determine if this is due to increasing radiative effects due There are several issues which have emerged from this to CO2, or increasing water vapour feedbacks or lapse-rate study which should be addressed in future work aimed at rec- changes. However, it is clear that it is not due to increased onciling model simulations and proxy data reconstructions albedo feedbacks, or cloud processes. of the early Eocene (many of which also apply to other time Given that the models prescribe Eocene vegetation in quite periods). different ways, it is interesting to assess how much this af- Firstly, modelling groups should aim to carry out simula- fects inter-model variability. Figure 12 shows the surface tions over a wider range of atmospheric CO2 levels. In partic- albedo in the pre-industrial control and the ×2 CO2 sim- ular, the results of CCSM H indicate that at high prescribed ulations for HadCM, CCSM H, and ECHAM. At the high atmospheric CO2 and low aerosol forcing, the models and latitudes, this is affected by snow and sea ice cover and data come close together. Some of this work is in progress prescribed changes in ice sheets, but at low latitudes this is (e.g. simulations at ×3 CO2 are currently being analysed for purely a result of the imposed vegetation and open-ocean the ECHAM model). However, it should be noted that this is albedos. The fact that all the models have a low latitude not always possible. For example, the Eocene HadCM model albedo which is similar to their control, and similar to each has been run at ×8 CO2, but after about 2700 yr the model other, indicates that this aspect of experimental design is developed a runaway greenhouse, and the model eventually likely not playing an important role in determining the dif- crashed (Lunt et al., 2007). A similar effect has been ob- ferences in results between the models. served in the ECHAM model at ×4 CO2 (Heinemann, 2009). Whether such an effect is “real”, i.e. whether the real world would also develop a runaway greenhouse, is completely un- 6 Conclusions and outlook known. In any case, it is clear that modelling the early Eocene climate pushes the climate model parameterisations to the We have carried out an intercomparison and model–data boundaries within which they were designed to operate, if comparison of the results from 5 early Eocene modelling not beyond these boundaries. studies, using 4 different climate models. The model results Some of the differences between the model results can show a large spread in global mean temperatures, for exam- be attributed to differences in the experimental design. In ple a ∼ 9 ◦C range in surface air temperature under a single particular, some models apply a very generic Eocene veg- etation, which is not particularly realistic. A slightly more CO2 value, and are characterised by warming in different re- gions. The models which have been run at sufficiently high coordinated study could provide guidelines for ways to bet- ter represent Eocene vegetation, for example by making use CO2 show very good agreement with the terrestrial data. The comparison with SST data is also good, but the model and of palynological data or by using dynamic vegetation mod- data uncertainty only just overlap for the Arctic and south- els where available. This would provide an ensemble of 18 model results which better represented the true uncertainty in west Pacific δ O and TEX86 proxies. However, if a possi- ble seasonality bias in the proxies is taken into account, then our model simulations. Other inconsistencies between model

www.clim-past.net/8/1717/2012/ Clim. Past, 8, 1717–1736, 2012 1734 D. J. Lunt et al.: EoMIP simulations should not necessarily be eliminated – for ex- Bice, K. L. and Marotzke, J.: Numerical evidence against reversed ample, different models using different paleogeographical thermohaline circulation in the warm Paleocene/Eocene ocean, reconstructions may be more representative of the true J. Geophys. Res., 106, 11529–11542, 2001. spread of model results than if all groups used a single Bijl, P. K., Schouten, S., Sluijs, A., Reichart, G. J., Zachos, J. C., paleogeography. and Brinkhuis, H.: Early Palaeogene temperature evolution of the On the data side, better understanding of the temperature southwest Pacific Ocean, Nature, 461, 776–779, 2009. Braconnot, P., Otto-Bliesner, B., Harrison, S., Joussaume, S., Pe- proxies and their associated uncertainties, in particular sea- terchmitt, J.-Y., Abe-Ouchi, A., Crucifix, M., Driesschaert, E., sonal effects, is a clear goal for future work, as is greater geo- Fichefet, Th., Hewitt, C. D., Kageyama, M., Kitoh, A., Laˆıne,´ graphical and finer temporal coverage. More tropical Eocene A., Loutre, M.-F., Marti, O., Merkel, U., Ramstein, G., Valdes, terrestrial data would be especially beneficial for assessing P., Weber, S. L., Yu, Y., and Zhao, Y.: Results of PMIP2 coupled the terrestrial meridional temperature gradient. simulations of the Mid-Holocene and Last Glacial Maximum – Perhaps most crucial of all, better CO2 constraints from Part 1: experiments and large-scale features, Clim. Past, 3, 261– proxies would be of huge benefit to model–data comparison 277, doi:10.5194/cp-3-261-2007, 2007. exercises such as this. Recently, much work is being under- Brinkhuis, H., Schouten, S., Collinson, M. E., Sluijs, A., Damste,´ taken in this area, but this should be intensified wherever pos- J. S. S., Dickens, G. R., Huber, M., Cronin, T. M., Onodera, J., Takahashi, K., Bujak, J. P., Stein, R., van der Burgh, J., sible. We note that at high CO2, due to the logarithmic nature of the CO forcing, proxies which may have relatively coarse Eldrett, J. S., Harding, I. C., Lotter, A. F., Sangiorgi, F., van 2 Konijnenburg-van Cittert, H., de Leeuw, J. W., Matthiessen, precision at low CO can actually provide very strong con- 2 J., Backman, J., Moran, K., and the Expedition 302 Scientists: straints on the CO2 forcing itself. Such constraints on CO2, Episodic fresh surface waters in the Eocene Arctic Ocean, Na- combined with proxy temperature reconstructions with well ture, 441, 606–609, 2006. defined uncertainty ranges, could provide a strong constraint Coggon, R. M., Teagle, D. A. H., Smith-Duque, C. E., Alt, J. C., and on model simulations, providing quantitative metrics for as- Cooper, M. J.: Reconstructing past seawater Mg/Ca and Sr/Ca sessing model performance, and could ultimately provide rel- from mid-ocean ridge flank calcium carbonate veins, Science, ative weightings for model simulations of future climates. 327, 1114–1117, 2010. Coggon, R. M., Teagle, D. A. H., and Dunkley Jones, T.: Comment on What do we know about the evolution of Mg to Ca ratios in Supplementary material related to this article is seawater? by Wally Broecker and Jimin Yu, Paleoceanography, available online at: http://www.clim-past.net/8/1717/ 26, PA3224, doi:10.1029/2011PA002186, 2011. 2012/cp-8-1717-2012-supplement.zip. Collins, W. D., Bitz, C. M., Blackmon, M. L., Bonan, G. B., Bretherton, C. S., Carton, J. A., Chang, P., Doney, S. C., Hack, J. J., Henderson, T. B., Kiehl, J. T., Large, W. G., Mckenna, D. S., Acknowledgements. This paper resulted from discussions at the Santer, B. D., and Smith, R. D.: The Community Climate System “CO2 at the zoo” event held in Bristol in 2010, and from a Royal Model Version 3 (CCSM3), J. Climate, 19, 2122–2143, 2006. Society Kavli meeting held in 2011. DJL acknowledges funding Colman, R. and McAveney, B.: Climate feedbacks under a very as an RCUK research fellow. NCEP Reanalysis Derived data pro- broad range of forcing, Geophysical Research Letters, 36, vided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, L01702, doi:10.1029/2008GL036268, 2009. from their Web site at http://www.esrl.noaa.gov/psd/. M. H. was Cox, P. M., Betts, R. A., Jones, C. D., Spall, S. A., and Totterdell, funded by NSF P2C2 grant OCE 0902882. This is PCCRC paper I. J.: Modelling vegetation and the carbon cycle as interactive ele- number 1220. ments of the climate system, in: Meteorology at the Millennium, edited by: Pearce, R., Academic Press, 259–279, 2001. Edited by: A. Haywood Cramer, B. S., Miller, K. G., Barrett, P. J., and Wright, J. D.: Late Cretaceous-Neogene trends in deep ocean temperature and con- tinental ice volume: Reconciling records of benthic foraminiferal References geochemistry (δ18O and Mg/Ca) with sea level history, J. Geo- phys. Res., 116, C12023, doi:10.1029/2011JC007255, 2011. Abbot, D. S. and Tziperman, E.: Sea Ice, High Latitude Convec- Creech, J. B., Baker, J. A., Hollis, C. J., Morgans, H. E. G., and tion, and Equable Climates, Geophysical Research Letters, 35, Smith, E. G. C.: Eocene sea temperatures for the mid-latitude L03702, doi:10.1029/2007GL032286, 2008. southwest Pacific from Mg/Ca ratios in planktonic and benthic Anand, P., Elderfield, H., and Conte, M.: Calibration of foraminifera, Earth Planet. Sci. Lett., 299, 483–495, 2010. Mg/Ca thermometry in planktonic foraminifera from a Douglas, P. M., Ivany, L., Pagani, M., Hollis, C. J., Beu, A. G., and sediment trap time series, Paleoceanography, 18, 1050, Affek, H. P.: Eocene Southern high latitude sea surface tempera- doi:10.1029/2002PA000846, 2003. tures: new constraints from clumped isotope paleothermometry, Beerling, D. and Royer, D.: Convergent Cenozoic CO2 history, Nat. AGU fall meeting, San Francisco, PP33B–1924, 2011. Geosci., 4, 418–420, 2011. Dowsett, H. J., Robinson, M. M., and Foley, K. M.: Pliocene three- Bemis, B., Spero, H., Bijma, J., and Lea, D.: Reevaluation of the dimensional global ocean temperature reconstruction, Clim. Past, oxygen isotopic composition of planktonic foraminifera: Exper- 5, 769–783, doi:10.5194/cp-5-769-2009, 2009. imental results and revised paleotemperature equations, Paleo- ceanography, 13, 150–160, 1998.

Clim. Past, 8, 1717–1736, 2012 www.clim-past.net/8/1717/2012/ D. J. Lunt et al.: EoMIP 1735

Farkas,ˇ J., B’ohm,´ F., Wallmann, K., Blenkinsop, J., Eisenhauer, A., Kobashi, T., Grossman, E., Yancey, T. E., and Dockery, D. T.: The van Geldern, R., Munnecke, A., Voigt, S., and Veizer, J.: Calcium oxygen isotopic record of seasonality in Conus shells and its ap- isotope record of Phanerozoic oceans: Implications for chemical plications to understanding the late middle Eocene (38 Ma) cli- evolution of seawater and its causative mechanisms, Geochem. mate, Paleontol. Res., 7, 343–355, 2003. Cosmochem. Acta, 71, 5117–5134, 2007. Kump, L. R. and Pollard, D.: Amplification of Cretaceous Warmth Grossman, E. and Ku, T. L.: Oxygen and carbon isotope fractiona- by Biological Cloud Feedbacks, Science, 320, p. 195, 2008. tion in biogenic aragonite: Temperature effects, Chem. Geol., 59, Lear, C., Rosenthal, Y., and Slowey, N.: Benthic foraminiferal 59–74, 1986. Mg/Ca-paleothermometry: A revised core-top calibration, Hegerl, G. C., Zwiers, F. W., Braconnot, P., Gillett, N. P., Luo, Y., Geochimica et Cosmochimica Acta, 66, 3375–3387, 2002. Orsini, J. A. M., Nicholls, N., Penner, J. E., and Stott, P. A.: Liu, Z., Pagani, M., Zinniker, D., DeConto, R. M., Huber, M., Climate Change 2007: The Physical Science Basis. Contribu- Brinkhuis, H., Shah, S. R., Leckie, R. M., and Pearson, A.: tion of Working Group I to the Fourth Assessment Report of Global Cooling During the Eocene-Oligocene Climate Transi- the Intergovernmental Panel on Climate Change, chap. Under- tion, Science, 323, 1187–1190, 2009. standing and Attributing Climate Change, Cambridge University Lunt, D. J., Valdes, P. J., and Ridgwell, A.: On the linearity Press, 2007. of Climate Sensitivity: A case study for the future and the Heinemann, M.: Warm and sensitive Paleocene-Eocene climate, PETM, AGU poster, available at: http://www.paleo.bris.ac.uk/ Max Planck Institute for Meteorology, Hamburg, Germany, ∼ggdjl/conferences/agu2007 climsens.pdf (last access: 18 Octo- Reports on Earth System Science 70, http://www.mpimet.mpg. ber 2012), 2007. de/fileadmin/staff/heinemannmalte/publications/MPI Rep70.pdf Lunt, D. J., Valdes, P. J., Dunkley Jones, T., Ridgwell, A., Hay- (last access: 18 October 2012), 2009. wood, A. M., Schmidt, D. N., Marsh, R., and Maslin, M.: CO2- Heinemann, M., Jungclaus, J. H., and Marotzke, J.: Warm Pale- driven ocean circulation changes as an amplifier of Paleocene- ocene/Eocene climate as simulated in ECHAM5/MPI-OM, Clim. Eocene thermal maximum hydrate destabilization, Geology, 38, Past, 5, 785–802, doi:10.5194/cp-5-785-2009, 2009. 875–878, 2010. Hollis, C. J., Handley, L., Crouch, E. M., Morgans, H. E. G., Baker, Meehl, G. A., Stocker, T. F., Collins, W. D., Friedlingstein, P., Gaye, J. A., Creech, J., Collins, K. S., Gibbs, S. J., Huber, M., Schouten, A. T., Gregory, J. M., Kitoh, A., Knutti, R., Murphy, J. M., Noda, S., Zachos, J. C., and Pancost, R. D.: Tropical sea temperatures in A., Raper, S. C. B., Watterson, I. G., Weaver, A. J., and Zhao, the high-latitude South Pacific during the Eocene, Geology, 37, Z.-C.: Climate Change 2007: The Physical Science Basis. Con- 99–102, 2009. tribution of Working Group I to the Fourth Assessment Report Hollis, C. J., Taylor, K. W. R., Handley, L., Pancost, R. D., Hu- of the Intergovernmental Panel on Climate Change, chap. Global ber, M., Creech, J. B., Hines, B. R., Crouch, E. M., Morgans, climate projections, Cambridge University Press, 2007. H. E. G., Crampton, J. S., Gibbs, S., Pearson, P. N., and Zachos, Moss, R. H., Edmonds, J. A., Hibbard, K. A., Manning, M. R., Rose, J. C.: Early Paleogene temperature history of the Southwest Pa- S. K., van Vuuren, D. P., Carter, T. R., Emori, S., Kainuma, M., cific Ocean: Reconciling proxies and models, Earth Planet. Sci. Kram, T., Meehl, G. A., Mitchell, J. F. B., Nakicenovic, N., Riahi, Lett., 349–350, 53–66, 2012. K., Smith, S. J., Stouffer, R. J., Thomson, A. M., Weyant, J. P., Huber, M. and Caballero, R.: The early Eocene equable climate and Wilbanks, T. J.: The next generation of scenarios for climate problem revisited, Clim. Past, 7, 603–633, doi:10.5194/cp-7- change research and assessment, Nature, 463, 747–756, 2010. 603-2011, 2011. Nicholas, C., Pearson, P. N., Bown, P. R., Dunkley Jones, T., Huber, Huber, M., Sloan, L. C., and Shellito, C.: Causes and consequences B. T., Karega, A., Lees, J. A., McMillan, I. K., O’Halloran, A., of globally warm climates in the early Paleogene, chap. Early Pa- Singano, J. M., and Wade, B. S.: Stratigraphy and sedimentol- leogene oceans and climate: A fully coupled modeling approach ogy of the Upper Cretaceous to Paleogene Kilwa Group, south- using the NCAR CCSM, Geological Society of America Special ern coastal Tanzania, J. Afr. Earth Sci., 45, 431–466, 2006. Paper 369, 2003. Pagani, M., Liu, Z., LaRiviere, L., and Ravelo, A. C.: High Earth- Ivany, L. C., Lohmann, K. C., Hasiuk, F., Blake, D. B., Glass, A., system climate sensitivity determined from Pliocene carbon Aronson, R. B., and Moody, R. M.: Eocene climate record of a dioxide concentrations, Nat. Geosci., 3, 27–30, 2010. high southern latitude continental shelf: Seymour Island, Antarc- Pagani, M., Huber, M., Liu, Z., Bohaty, S. M., Henderiks, J., Sijp, tica, Geol. Soc. Am. Bull., 150, 659–678, 2008. W., Krishnan, S., and DeConto, R. M.: The role of carbon diox- Kalnay, E., Kanamitsu, M., Kistler, R., Collins, W., Deaven, D., ide during the onset of Antarctic glaciation, Science, 334, 1261– Gandin, L., Iredell, M., Saha, S., White, G., Woollen, J., Zhu, Y., 1264, 2011. Leetmaa, A., Reynolds, R., Chelliah, M., Ebisuzaki, W., Higgins, Panchuk, K., Ridgwell, A., and Kump, L. R.: Sedimentary re- W., Janowiak, J., Mo, K. C., Ropelewski, C., Wang, J., Jenne, sponse to Paleocene-Eocene Thermal Maximum carbon release: R., and Joseph, D.: The NCEP/NCAR 40-year reanalysis project, A model-data comparison, Geology, 36, 315–318, 2008. Bull. Amer. Meteor. Soc., 77, 437–470, 1996. Payros, A., Bernaola, G., Orue-Etxebarria, X., Tosquella, J. D.-T., Kim, J.-H., Meer, J. V. D., Schouten, S., Helmke, P., Willmott, V., and Apellaniz, E.: Reassessment of the Early-Middle Eocene Sangiorgi, F., Koc¸, N., Hopmans, E. C., and Damste,´ J. S. S.: New biomagnetochronology based on evidence from the Gorrondaxte indices and calibrations derived from the distribution of crenar- section (Basque Country, western Pyrenees), Lethaia, 40, 183– chaeal isoprenoid tetraether lipids: Implications for past sea sur- 195, 2007. face temperature reconstructions, Geochim. Cosmochim. Acta, Pearson, P. N., Ditchfield, P. W., Singano, J., Harcourt-Brown, 74, 4639–4654, 2010. K. G., Nicholas, C. J., and Olsson, R. K.: Warm tropical sea surface temperatures in the Late Cretaceous and Eocene epochs,

www.clim-past.net/8/1717/2012/ Clim. Past, 8, 1717–1736, 2012 1736 D. J. Lunt et al.: EoMIP

Nature, 413, 481–487, 2001. Seki, O., Foster, G. L., Schmidt, D. N., Mackenden, A., Kawamura, Pearson, P. N., Nicholas, C. J., Singano, J. M., Bown, P. R., Cox- K., and Pancost, R. D.: Alkenone and boron-based Pliocene all, H. K., van Dongen, B. E., Huber, B. T., Karega, A., amd pCO2 records, Earth Planet. Sci. Lett., 292, 201–211, 2010. E. Msaky, J. A. L., Pancost, R. D., Pearson, M., and Roberts, Sewall, J. O., Sloan, L. C., Huber, M., and Wing, S.: Climate sen- A. P.: Paleogene and cretaceous sediment cores from the Kilwa sitivity to changes in land surface characteristics, Glob. Planet. and Lindi areas of coastal Tanzania: Tanzania Drilling Project Change, 26, 445–465, 2000. Sites 1–5, J. Afr. Earth Sci., 39, 25–62, 2004. Sexton, P. F., Wilson, P. A., and Pearson, P. N.: Microstructural and Pearson, P. N., van Dongen, B. E., Nicholas, C. J., Pancost, R. D., geochemical perspectives on planktic foraminiferal preservation: Schouten, S., Singano, J. M., and Wade, B. S.: Stable warm trop- Glassy versus Frosty, Geochem. Geophys. Geosyst., 7, Q12P19, ical climate through the Eocene Epoch, Geology, 35, 211–214, doi:10.1029/2006GC001291, 2006. 2007. Shellito, C. J. and Sloan, L. C.: Reconstructing a lost Eocene par- Pearson, P. N., Foster, G. L., and Wade, B. S.: Atmospheric car- adise: Part I. Simulating the change in global floral distribution at bon dioxide through the Eocene-Oligocene climate transition, the initial Eocene thermal maximum, Glob. Planet. Change, 50, Nature, 461, 1110–1113, 2009. 1–17, 2006. Rathmann, S. and Kuhnert, H.: Carbonate ion effect on Mg/Ca, Sluijs, A., Schouten, S., Pagani, M., Woltering, M., Brinkhuis, H., Sr/Ca and stable isotopes on the benthic foraminifera Oridor- Damste,´ J. S., Dickens, G. R., Huber, M., Reichart, G. J., Stein, salis umbonatus off Namibia, Mar. Micropaleontol., 66, 120– R., an L. J. Lourens, J. M., Pedentchouk, N., Backman, J., and 133, 2008. Moran, K.: Subtropical Arctic Ocean temperatures during the Rathmann, S., Hess, S., Kuhnert, H., and Mulitza, S.: Mg/Ca ratios Palaeocene/Eocene thermal maximum, Nature, 441, 610–613, of the benthic foraminifera Oridorsalis umbonatus obtained by 2006. laser ablation from core top sediments: Relationship to bottom Sluijs, A., Rohl, U., Schouten, S., Sangiorgi, H. J. B. F., Sin- water temperatures, Geochem., Geophys., Geosyst., 5, Q12013, ninghe Damste,´ J. S., and Brinkhuis, H.: Arctic late Paleocene- doi:10.1029/2004GC000808, 2004. early Eocene paleoenvironments with special emphasis on the Roberts, C. D., LeGrande, A. N., and Tripati, A. K.: Climate sen- Paleocene-Eocene thermal maximum (Lomonosov Ridge, Inte- sitivity to Arctic seaway restriction during the early Paleogene, grated Ocean Drilling Program Expedition 302), Paleoceanogra- Earth Planet. Science Lett., 286, 576–585, 2009. phy, 23, PA1S11, doi:10.1029/2007PA001495, 2008. Roberts, C. D., LeGrande, A. N., and Tripati, A. K.: Sensitivity of Storey, M., Duncan, R. A., and Swisher, C. C.: Paleocene-Eocene seawater oxygen isotopes to climatic and tectonic boundary con- thermal maximum and the opening of the northeast Atlantic, Sci- ditions in an early Paleogene simulation with GISS ModelE-R, ence, 316, 587–589, 2007. Paleoceanography, 126, PA4203, doi:10.1029/2010PA002025, Tindall, J., Flecker, R., Valdes, P., Schmidt, D. N., Markwick, P., 2011. and Harris, J.: Modelling oxygen isotope distribution of ancient Roeckner, E., Bauml, G., Bonaventura, L., Brokopf, R., Esch, seawater using a coupled ocean-atmosphere GCM: Implications M., Giorgetta, M., Hagemann, S., Kirchner, I., Kornblueh, for reconstructing early Eocene climate, Earth Planet. Sci. Lett., L., Manzini, E., Rhodin, A., Schlese, U., Schulzweida, U., 292, 265–273, 2010. and Tompkins, A.: The atmospheric general circulation model Winguth, A., Shellito, C., Shields, C., and Winguth, C.: Climate ECHAM5, Tech. rep., Max Planck Institute for Meteorology, Response at the Paleocene-Eocene Thermal Maximum to Green- Hamburg, Germany, available at: http://www.mpimet.mpg.de house Gas Forcing – A Model Study with CCSM3, J. Climate, (last access: 18 October 2012), 2003. 23, 2562–2584, 2010. Schmidt, G. A., Ruedy, R., Hansen, J. E., Aleinov, I., Bell, N., Winguth, A. M. E., Thomas, E., and Winguth, C.: Global decline Bauer, M., Bauer, S., Cairns, B., Canuto, V., Cheng, Y., Genio, in ocean ventilation, oxygenation, and productivity during the A. D., Faluvegi, G., Friend, A. D., Hall, T. M., Hu, Y., Kelley, Paleocene-Eocene Thermal Maximum: Implications for the ben- M., Kiang, N. Y., Koch, D., Lacis, A. A., Lerner, J., Lo, K. K., thic extinction, Geology, 40, 263–266, doi:10.1130/G32529.1, Miller, R. L., Nazarenko, L., Oinas, V., Perlwitz, J. P., Perlwitz, 2012. J., Rind, D., Romanou, A., Russell, G. L., Sato, M., Shindell, Yeager, S. G., Shields, C. A., Large, W. G., and Hack, J. J.: The D. T., Stone, P. H., Sun, S., Tausnev, N., Thresher, D., , and Yao, low-resolution CCSM3, J. Climate, 19, 2545–2566, 2006. M. S.: Present day atmospheric simulations using GISS ModelE: Zachos, J. C., Dickens, G. R., and Zeebe, R. E.: An early Cenozoic Comparison to in-situ, satellite and reanalysis data, J. Climate, perspective on greenhouse warming and carbon-cycle dynamics, 19, 153–192, 2006. Nature, 451, 279–283, 2008.

Clim. Past, 8, 1717–1736, 2012 www.clim-past.net/8/1717/2012/